@tensorflow/tfjs-layers

@tensorflow/tfjs-layers

Version 4.22.0
Published 2 months ago
31.1 MB
No dependencies
Apache-2.0 AND MIT license

Install

npm i @tensorflow/tfjs-layers

yarn add @tensorflow/tfjs-layers

pnpm add @tensorflow/tfjs-layers

Overview

TensorFlow layers API in JavaScript

Index

Variables

variable callbacks

const callbacks: { earlyStopping: typeof earlyStopping };

Functions

function input

input: (config: InputConfig) => SymbolicTensor;

Used to instantiate an input to a model as a tf.SymbolicTensor.
Users should call the input factory function for consistency with other generator functions.
Example:
// Defines a simple logistic regression model with 32 dimensional input // and 3 dimensional output. const x = tf.input({shape: [32]}); const y = tf.layers.dense({units: 3, activation: 'softmax'}).apply(x); const model = tf.model({inputs: x, outputs: y}); model.predict(tf.ones([2, 32])).print();
Note: input is only necessary when using model. When using sequential, specify inputShape for the first layer or use inputLayer as the first layer.
{heading: 'Models', subheading: 'Inputs'}

function loadLayersModel

loadLayersModel: (
    pathOrIOHandler: string | io.IOHandler,
    options?: io.LoadOptions
) => Promise<LayersModel>;

Load a model composed of Layer objects, including its topology and optionally weights. See the Tutorial named "How to import a Keras Model" for usage examples.
This method is applicable to:
1. Models created with the tf.layers.*, tf.sequential, and tf.model APIs of TensorFlow.js and later saved with the tf.LayersModel.save method. 2. Models converted from Keras or TensorFlow tf.keras using the [tensorflowjs_converter](https://github.com/tensorflow/tfjs/tree/master/tfjs-converter).
This mode is *not* applicable to TensorFlow SavedModels or their converted forms. For those models, use tf.loadGraphModel.
Example 1. Load a model from an HTTP server.
const model = await tf.loadLayersModel( 'https://storage.googleapis.com/tfjs-models/tfjs/iris_v1/model.json'); model.summary();
Example 2: Save model's topology and weights to browser [local storage](https://developer.mozilla.org/en-US/docs/Web/API/Window/localStorage); then load it back.
const model = tf.sequential( {layers: [tf.layers.dense({units: 1, inputShape: [3]})]}); console.log('Prediction from original model:'); model.predict(tf.ones([1, 3])).print(); const saveResults = await model.save('localstorage://my-model-1'); const loadedModel = await tf.loadLayersModel('localstorage://my-model-1'); console.log('Prediction from loaded model:'); loadedModel.predict(tf.ones([1, 3])).print();
Example 3. Saving model's topology and weights to browser [IndexedDB](https://developer.mozilla.org/en-US/docs/Web/API/IndexedDB_API); then load it back.
const model = tf.sequential( {layers: [tf.layers.dense({units: 1, inputShape: [3]})]}); console.log('Prediction from original model:'); model.predict(tf.ones([1, 3])).print(); const saveResults = await model.save('indexeddb://my-model-1'); const loadedModel = await tf.loadLayersModel('indexeddb://my-model-1'); console.log('Prediction from loaded model:'); loadedModel.predict(tf.ones([1, 3])).print();
Example 4. Load a model from user-selected files from HTML [file input elements](https://developer.mozilla.org/en-US/docs/Web/HTML/Element/input/file).
// Note: this code snippet will not work without the HTML elements in the // page const jsonUpload = document.getElementById('json-upload'); const weightsUpload = document.getElementById('weights-upload'); const model = await tf.loadLayersModel( tf.io.browserFiles([jsonUpload.files[0], weightsUpload.files[0]]));
Parameter pathOrIOHandler
Can be either of the two formats 1. A string path to the ModelAndWeightsConfig JSON describing the model in the canonical TensorFlow.js format. For file:// (tfjs-node-only), http:// and https:// schemas, the path can be either absolute or relative. The content of the JSON file is assumed to be a JSON object with the following fields and values: - 'modelTopology': A JSON object that can be either of: 1. a model architecture JSON consistent with the format of the return value of keras.Model.to_json() 2. a full model JSON in the format of keras.models.save_model(). - 'weightsManifest': A TensorFlow.js weights manifest. See the Python converter function save_model() for more details. It is also assumed that model weights can be accessed from relative paths described by the paths fields in weights manifest. 2. A tf.io.IOHandler object that loads model artifacts with its load method.
Parameter options
Optional configuration arguments for the model loading, including: - strict: Require that the provided weights exactly match those required by the layers. Default true. Passing false means that both extra weights and missing weights will be silently ignored. - onProgress: A progress callback of the form: (fraction: number) => void. This callback can be used to monitor the model-loading process.
Returns
A Promise of tf.LayersModel, with the topology and weights loaded.
{heading: 'Models', subheading: 'Loading'}

function model

model: (args: ContainerArgs) => LayersModel;

A model is a data structure that consists of Layers and defines inputs and outputs.

The key difference between tf.model and tf.sequential is that tf.model is more generic, supporting an arbitrary graph (without cycles) of layers. tf.sequential is less generic and supports only a linear stack of layers.

When creating a tf.LayersModel, specify its input(s) and output(s). Layers are used to wire input(s) to output(s).

For example, the following code snippet defines a model consisting of two dense layers, with 10 and 4 units, respectively.

// Define input, which has a size of 5 (not including batch dimension).
const input = tf.input({shape: [5]});

// First dense layer uses relu activation.
const denseLayer1 = tf.layers.dense({units: 10, activation: 'relu'});
// Second dense layer uses softmax activation.
const denseLayer2 = tf.layers.dense({units: 4, activation: 'softmax'});

// Obtain the output symbolic tensor by applying the layers on the input.
const output = denseLayer2.apply(denseLayer1.apply(input));

// Create the model based on the inputs.
const model = tf.model({inputs: input, outputs: output});

// The model can be used for training, evaluation and prediction.
// For example, the following line runs prediction with the model on
// some fake data.
model.predict(tf.ones([2, 5])).print();

See also: tf.sequential, tf.loadLayersModel.

{heading: 'Models', subheading: 'Creation'}

function registerCallbackConstructor

registerCallbackConstructor: (
    verbosityLevel: number,
    callbackConstructor: BaseCallbackConstructor
) => void;

function sequential

sequential: (config?: SequentialArgs) => Sequential;

Creates a tf.Sequential model. A sequential model is any model where the outputs of one layer are the inputs to the next layer, i.e. the model topology is a simple 'stack' of layers, with no branching or skipping.

This means that the first layer passed to a tf.Sequential model should have a defined input shape. What that means is that it should have received an inputShape or batchInputShape argument, or for some type of layers (recurrent, Dense...) an inputDim argument.

The key difference between tf.model and tf.sequential is that tf.sequential is less generic, supporting only a linear stack of layers. tf.model is more generic and supports an arbitrary graph (without cycles) of layers.

Examples:

const model = tf.sequential();

// First layer must have an input shape defined.
model.add(tf.layers.dense({units: 32, inputShape: [50]}));
// Afterwards, TF.js does automatic shape inference.
model.add(tf.layers.dense({units: 4}));

// Inspect the inferred shape of the model's output, which equals
// `[null, 4]`. The 1st dimension is the undetermined batch dimension; the
// 2nd is the output size of the model's last layer.
console.log(JSON.stringify(model.outputs[0].shape));

It is also possible to specify a batch size (with potentially undetermined batch dimension, denoted by "null") for the first layer using the batchInputShape key. The following example is equivalent to the above:

const model = tf.sequential();

// First layer must have a defined input shape
model.add(tf.layers.dense({units: 32, batchInputShape: [null, 50]}));
// Afterwards, TF.js does automatic shape inference.
model.add(tf.layers.dense({units: 4}));

// Inspect the inferred shape of the model's output.
console.log(JSON.stringify(model.outputs[0].shape));

You can also use an Array of already-constructed Layers to create a tf.Sequential model:

const model = tf.sequential({
  layers: [tf.layers.dense({units: 32, inputShape: [50]}),
           tf.layers.dense({units: 4})]
});
console.log(JSON.stringify(model.outputs[0].shape));

{heading: 'Models', subheading: 'Creation'}

Classes

class Callback

abstract class Callback extends BaseCallback {}

property model

model: LayersModel;

Instance of keras.models.Model. Reference of the model being trained.

method setModel

setModel: (model: Container) => void;

class CallbackList

class CallbackList {}

Container abstracting a list of callbacks.

constructor

constructor(callbacks?: BaseCallback[], queueLength?: number);

Constructor of CallbackList.
Parameter callbacks
Array of Callback instances.
Parameter queueLength
Queue length for keeping running statistics over callback execution time.

property callbacks

callbacks: BaseCallback[];

property queueLength

queueLength: number;

method append

append: (callback: BaseCallback) => void;

method onBatchBegin

onBatchBegin: (batch: number, logs?: UnresolvedLogs) => Promise<void>;

Called right before processing a batch.
Parameter batch
Index of batch within the current epoch.
Parameter logs
Dictionary of logs.

method onBatchEnd

onBatchEnd: (batch: number, logs?: UnresolvedLogs) => Promise<void>;

Called at the end of a batch.
Parameter batch
Index of batch within the current epoch.
Parameter logs
Dictionary of logs.

method onEpochBegin

onEpochBegin: (epoch: number, logs?: UnresolvedLogs) => Promise<void>;

Called at the start of an epoch.
Parameter epoch
Index of epoch.
Parameter logs
Dictionary of logs.

method onEpochEnd

onEpochEnd: (epoch: number, logs?: UnresolvedLogs) => Promise<void>;

Called at the end of an epoch.
Parameter epoch
Index of epoch.
Parameter logs
Dictionary of logs.

method onTrainBegin

onTrainBegin: (logs?: UnresolvedLogs) => Promise<void>;

Called at the beginning of training.
Parameter logs
Dictionary of logs.

method onTrainEnd

onTrainEnd: (logs?: UnresolvedLogs) => Promise<void>;

Called at the end of training.
Parameter logs
Dictionary of logs.

method setModel

setModel: (model: Container) => void;

method setParams

setParams: (params: Params) => void;

class CustomCallback

class CustomCallback extends BaseCallback {}

Custom callback for training.

constructor

constructor(args: CustomCallbackArgs, yieldEvery?: YieldEveryOptions);

property batchBegin

protected readonly batchBegin: (
    batch: number,
    logs?: Logs
) => void | Promise<void>;

property batchEnd

protected readonly batchEnd: (
    batch: number,
    logs?: Logs
) => void | Promise<void>;

property epochBegin

protected readonly epochBegin: (
    epoch: number,
    logs?: Logs
) => void | Promise<void>;

property epochEnd

protected readonly epochEnd: (
    epoch: number,
    logs?: Logs
) => void | Promise<void>;

property nextFrameFunc

nextFrameFunc: Function;

property nowFunc

nowFunc: Function;

property trainBegin

protected readonly trainBegin: (logs?: Logs) => void | Promise<void>;

property trainEnd

protected readonly trainEnd: (logs?: Logs) => void | Promise<void>;

property yield

protected readonly yield: (
    epoch: number,
    batch: number,
    logs: Logs
) => void | Promise<void>;

method maybeWait

maybeWait: (epoch: number, batch: number, logs: UnresolvedLogs) => Promise<void>;

method onBatchBegin

onBatchBegin: (batch: number, logs?: UnresolvedLogs) => Promise<void>;

method onBatchEnd

onBatchEnd: (batch: number, logs?: UnresolvedLogs) => Promise<void>;

method onEpochBegin

onEpochBegin: (epoch: number, logs?: UnresolvedLogs) => Promise<void>;

method onEpochEnd

onEpochEnd: (epoch: number, logs?: UnresolvedLogs) => Promise<void>;

method onTrainBegin

onTrainBegin: (logs?: UnresolvedLogs) => Promise<void>;

method onTrainEnd

onTrainEnd: (logs?: UnresolvedLogs) => Promise<void>;

class EarlyStopping

class EarlyStopping extends Callback {}

A Callback that stops training when a monitored quantity has stopped improving.

constructor

constructor(args?: EarlyStoppingCallbackArgs);

property baseline

protected readonly baseline: number;

property minDelta

protected readonly minDelta: number;

property mode

protected readonly mode: 'auto' | 'min' | 'max';

property monitor

protected readonly monitor: string;

property monitorFunc

protected monitorFunc: (currVal: number, prevVal: number) => boolean;

property patience

protected readonly patience: number;

property verbose

protected readonly verbose: number;

method onEpochEnd

onEpochEnd: (epoch: number, logs?: Logs) => Promise<void>;

method onTrainBegin

onTrainBegin: (logs?: Logs) => Promise<void>;

method onTrainEnd

onTrainEnd: (logs?: Logs) => Promise<void>;

class History

class History extends BaseCallback {}

Callback that records events into a History object. This callback is automatically applied to every TF.js Layers model. The History object gets returned by the fit method of models.

property epoch

epoch: number[];

property history

history: { [key: string]: any[] };

method onEpochEnd

onEpochEnd: (epoch: number, logs?: UnresolvedLogs) => Promise<void>;

method onTrainBegin

onTrainBegin: (logs?: UnresolvedLogs) => Promise<void>;

method syncData

syncData: () => Promise<void>;

Await the values of all losses and metrics.

class InputSpec

class InputSpec {}

Specifies the ndim, dtype and shape of every input to a layer.
Every layer should expose (if appropriate) an inputSpec attribute: a list of instances of InputSpec (one per input tensor).
A null entry in a shape is compatible with any dimension, a null shape is compatible with any shape.

constructor

constructor(args: InputSpecArgs);

property axes

axes?: { [axis: number]: number };

Dictionary mapping integer axes to a specific dimension value.

property dtype

dtype?: DataType;

Expected datatype of the input.

property maxNDim

maxNDim?: number;

Maximum rank of the input.

property minNDim

minNDim?: number;

Minimum rank of the input.

property ndim

ndim?: number;

Expected rank of the input.

property shape

shape?: Shape;

Expected shape of the input (may include null for unchecked axes).

class LayersModel

class LayersModel extends Container implements tfc.InferenceModel {}

A tf.LayersModel is a directed, acyclic graph of tf.Layers plus methods for training, evaluation, prediction and saving.
tf.LayersModel is the basic unit of training, inference and evaluation in TensorFlow.js. To create a tf.LayersModel, use tf.LayersModel.
See also: tf.Sequential, tf.loadLayersModel.
{heading: 'Models', subheading: 'Classes'}

constructor

constructor(args: ContainerArgs);

property className

static className: string;

property history

history: History;

property isOptimizerOwned

protected isOptimizerOwned: boolean;

property isTraining

protected isTraining: boolean;

property loss

loss:
    | string
    | string[]
    | { [outputName: string]: string }
    | LossOrMetricFn
    | LossOrMetricFn[]
    | { [outputName: string]: LossOrMetricFn };

property lossFunctions

lossFunctions: LossOrMetricFn[];

property metrics

metrics:
    | string
    | LossOrMetricFn
    | (string | LossOrMetricFn)[]
    | { [outputName: string]: string | LossOrMetricFn };

property metricsNames

metricsNames: string[];

property metricsTensors

metricsTensors: [LossOrMetricFn, number][];

property optimizer

optimizer: Optimizer;

property optimizer_

protected optimizer_: Optimizer;

property stopTraining

stopTraining: boolean;

property stopTraining_

protected stopTraining_: boolean;

method checkTrainableWeightsConsistency

protected checkTrainableWeightsConsistency: () => void;

Check trainable weights count consistency.
This will raise a warning if this.trainableWeights and this.collectedTrainableWeights are inconsistent (i.e., have different numbers of parameters). Inconsistency will typically arise when one modifies model.trainable without calling model.compile() again.

method compile

compile: (args: ModelCompileArgs) => void;

Configures and prepares the model for training and evaluation. Compiling outfits the model with an optimizer, loss, and/or metrics. Calling fit or evaluate on an un-compiled model will throw an error.
Parameter args
a ModelCompileArgs specifying the loss, optimizer, and metrics to be used for fitting and evaluating this model.
{heading: 'Models', subheading: 'Classes'}

method dispose

dispose: () => DisposeResult;

method evaluate

evaluate: (
    x: Tensor | Tensor[],
    y: Tensor | Tensor[],
    args?: ModelEvaluateArgs
) => Scalar | Scalar[];

Returns the loss value & metrics values for the model in test mode.
Loss and metrics are specified during compile(), which needs to happen before calls to evaluate().
Computation is done in batches.
const model = tf.sequential({ layers: [tf.layers.dense({units: 1, inputShape: [10]})] }); model.compile({optimizer: 'sgd', loss: 'meanSquaredError'}); const result = model.evaluate( tf.ones([8, 10]), tf.ones([8, 1]), {batchSize: 4}); result.print();
Parameter x
tf.Tensor of test data, or an Array of tf.Tensors if the model has multiple inputs.
Parameter y
tf.Tensor of target data, or an Array of tf.Tensors if the model has multiple outputs.
Parameter args
A ModelEvaluateArgs, containing optional fields.
Scalar test loss (if the model has a single output and no metrics) or Array of Scalars (if the model has multiple outputs and/or metrics). The attribute model.metricsNames will give you the display labels for the scalar outputs.
{heading: 'Models', subheading: 'Classes'}

method evaluateDataset

evaluateDataset: (
    dataset: Dataset<{}>,
    args?: ModelEvaluateDatasetArgs
) => Promise<Scalar | Scalar[]>;

Evaluate model using a dataset object.
Note: Unlike evaluate(), this method is asynchronous (async).
Parameter dataset
A dataset object. Its iterator() method is expected to generate a dataset iterator object, the next() method of which is expected to produce data batches for evaluation. The return value of the next() call ought to contain a boolean done field and a value field. The value field is expected to be an array of two tf.Tensors or an array of two nested tf.Tensor structures. The former case is for models with exactly one input and one output (e.g. a sequential model). The latter case is for models with multiple inputs and/or multiple outputs. Of the two items in the array, the first is the input feature(s) and the second is the output target(s).
Parameter args
A configuration object for the dataset-based evaluation.
Returns
Loss and metric values as an Array of Scalar objects.
{heading: 'Models', subheading: 'Classes'}

method execute

execute: (
    inputs: Tensor | Tensor[] | NamedTensorMap,
    outputs: string | string[]
) => Tensor | Tensor[];

Execute internal tensors of the model with input data feed.
Parameter inputs
Input data feed. Must match the inputs of the model.
Parameter outputs
Names of the output tensors to be fetched. Must match names of the SymbolicTensors that belong to the graph.
Returns
Fetched values for outputs.

method fit

fit: (x: any, y: any, args?: ModelFitArgs) => Promise<History>;

Trains the model for a fixed number of epochs (iterations on a dataset).
const model = tf.sequential({ layers: [tf.layers.dense({units: 1, inputShape: [10]})] }); model.compile({optimizer: 'sgd', loss: 'meanSquaredError'}); for (let i = 1; i < 5 ; ++i) { const h = await model.fit(tf.ones([8, 10]), tf.ones([8, 1]), { batchSize: 4, epochs: 3 }); console.log("Loss after Epoch " + i + " : " + h.history.loss[0]); }
Parameter x
tf.Tensor of training data, or an array of tf.Tensors if the model has multiple inputs. If all inputs in the model are named, you can also pass a dictionary mapping input names to tf.Tensors.
Parameter y
tf.Tensor of target (label) data, or an array of tf.Tensors if the model has multiple outputs. If all outputs in the model are named, you can also pass a dictionary mapping output names to tf.Tensors.
Parameter args
A ModelFitArgs, containing optional fields.
A History instance. Its history attribute contains all information collected during training.
ValueError In case of mismatch between the provided input data and what the model expects.
{heading: 'Models', subheading: 'Classes'}

method fitDataset

fitDataset: <T>(
    dataset: Dataset<T>,
    args: ModelFitDatasetArgs<T>
) => Promise<History>;

Trains the model using a dataset object.
Parameter dataset
A dataset object. Its iterator() method is expected to generate a dataset iterator object, the next() method of which is expected to produce data batches for training. The return value of the next() call ought to contain a boolean done field and a value field. The value field is expected to be an array of two tf.Tensors or an array of two nested tf.Tensor structures. The former case is for models with exactly one input and one output (e.g. a sequential model). The latter case is for models with multiple inputs and/or multiple outputs. Of the two items in the array, the first is the input feature(s) and the second is the output target(s).
Parameter args
A ModelFitDatasetArgs, containing optional fields.
A History instance. Its history attribute contains all information collected during training.
{heading: 'Models', subheading: 'Classes'}

method fitLoop

fitLoop: (
    f: (data: Tensor[]) => Scalar[],
    ins: Tensor[],
    outLabels?: string[],
    batchSize?: number,
    epochs?: number,
    verbose?: number,
    callbacks?: BaseCallback[],
    valF?: (data: Tensor[]) => Scalar[],
    valIns?: Tensor[],
    shuffle?: boolean | string,
    callbackMetrics?: string[],
    initialEpoch?: number,
    stepsPerEpoch?: number,
    validationSteps?: number
) => Promise<History>;

Abstract fit function for f(ins).
Parameter f
A Function returning a list of tensors. For training, this function is expected to perform the updates to the variables.
Parameter ins
List of tensors to be fed to f.
Parameter outLabels
List of strings, display names of the outputs of f.
Parameter batchSize
Integer batch size or == null if unknown. Default : 32.
Parameter epochs
Number of times to iterate over the data. Default : 1.
Parameter verbose
Verbosity mode: 0, 1, or 2. Default: 1.
Parameter callbacks
List of callbacks to be called during training.
Parameter valF
Function to call for validation.
Parameter valIns
List of tensors to be fed to valF.
Parameter shuffle
Whether to shuffle the data at the beginning of every epoch. Default : true.
Parameter callbackMetrics
List of strings, the display names of the metrics passed to the callbacks. They should be the concatenation of the display names of the outputs of f and the list of display names of the outputs of valF.
Parameter initialEpoch
Epoch at which to start training (useful for resuming a previous training run). Default : 0.
Parameter stepsPerEpoch
Total number of steps (batches on samples) before declaring one epoch finished and starting the next epoch. Ignored with the default value of undefined or null.
Parameter validationSteps
Number of steps to run validation for (only if doing validation from data tensors). Not applicable for tfjs-layers.
Returns
A History object.

method getDedupedMetricsNames

protected getDedupedMetricsNames: () => string[];

method getNamedWeights

protected getNamedWeights: (config?: io.SaveConfig) => NamedTensor[];

Extract weight values of the model.
Parameter config
: An instance of io.SaveConfig, which specifies model-saving options such as whether only trainable weights are to be saved.
Returns
A NamedTensorMap mapping original weight names (i.e., non-uniqueified weight names) to their values.

method getTrainingConfig

protected getTrainingConfig: () => TrainingConfig;

method getUserDefinedMetadata

getUserDefinedMetadata: () => {};

Get user-defined metadata.
The metadata is supplied via one of the two routes: 1. By calling setUserDefinedMetadata(). 2. Loaded during model loading (if the model is constructed via tf.loadLayersModel().)
If no user-defined metadata is available from either of the two routes, this function will return undefined.

method loadTrainingConfig

loadTrainingConfig: (trainingConfig: TrainingConfig) => void;

method makeTrainFunction

protected makeTrainFunction: () => (data: Tensor[]) => Scalar[];

Creates a function that performs the following actions:
1. computes the losses 2. sums them to get the total loss 3. call the optimizer computes the gradients of the LayersModel's trainable weights w.r.t. the total loss and update the variables 4. calculates the metrics 5. returns the values of the losses and metrics.

method predict

predict: (x: Tensor | Tensor[], args?: ModelPredictArgs) => Tensor | Tensor[];

Generates output predictions for the input samples.
Computation is done in batches.
Note: the "step" mode of predict() is currently not supported. This is because the TensorFlow.js core backend is imperative only.
const model = tf.sequential({ layers: [tf.layers.dense({units: 1, inputShape: [10]})] }); model.predict(tf.ones([8, 10]), {batchSize: 4}).print();
Parameter x
The input data, as a Tensor, or an Array of tf.Tensors if the model has multiple inputs.
Parameter args
A ModelPredictArgs object containing optional fields.
Prediction results as a tf.Tensor(s).
ValueError In case of mismatch between the provided input data and the model's expectations, or in case a stateful model receives a number of samples that is not a multiple of the batch size.
{heading: 'Models', subheading: 'Classes'}

method predictOnBatch

predictOnBatch: (x: Tensor | Tensor[]) => Tensor | Tensor[];

Returns predictions for a single batch of samples.
const model = tf.sequential({ layers: [tf.layers.dense({units: 1, inputShape: [10]})] }); model.predictOnBatch(tf.ones([8, 10])).print();
Parameter x
: Input samples, as a Tensor (for models with exactly one input) or an array of Tensors (for models with more than one input). Tensor(s) of predictions
{heading: 'Models', subheading: 'Classes'}

method save

save: (
    handlerOrURL: io.IOHandler | string,
    config?: io.SaveConfig
) => Promise<io.SaveResult>;

Save the configuration and/or weights of the LayersModel.
An IOHandler is an object that has a save method of the proper signature defined. The save method manages the storing or transmission of serialized data ("artifacts") that represent the model's topology and weights onto or via a specific medium, such as file downloads, local storage, IndexedDB in the web browser and HTTP requests to a server. TensorFlow.js provides IOHandler implementations for a number of frequently used saving mediums, such as tf.io.browserDownloads and tf.io.browserLocalStorage. See tf.io for more details.
This method also allows you to refer to certain types of IOHandlers as URL-like string shortcuts, such as 'localstorage://' and 'indexeddb://'.
Example 1: Save model's topology and weights to browser [local storage](https://developer.mozilla.org/en-US/docs/Web/API/Window/localStorage); then load it back.
const model = tf.sequential( {layers: [tf.layers.dense({units: 1, inputShape: [3]})]}); console.log('Prediction from original model:'); model.predict(tf.ones([1, 3])).print(); const saveResults = await model.save('localstorage://my-model-1'); const loadedModel = await tf.loadLayersModel('localstorage://my-model-1'); console.log('Prediction from loaded model:'); loadedModel.predict(tf.ones([1, 3])).print();
Example 2. Saving model's topology and weights to browser [IndexedDB](https://developer.mozilla.org/en-US/docs/Web/API/IndexedDB_API); then load it back.
const model = tf.sequential( {layers: [tf.layers.dense({units: 1, inputShape: [3]})]}); console.log('Prediction from original model:'); model.predict(tf.ones([1, 3])).print(); const saveResults = await model.save('indexeddb://my-model-1'); const loadedModel = await tf.loadLayersModel('indexeddb://my-model-1'); console.log('Prediction from loaded model:'); loadedModel.predict(tf.ones([1, 3])).print();
Example 3. Saving model's topology and weights as two files (my-model-1.json and my-model-1.weights.bin) downloaded from browser.
const model = tf.sequential( {layers: [tf.layers.dense({units: 1, inputShape: [3]})]}); const saveResults = await model.save('downloads://my-model-1');
Example 4. Send model's topology and weights to an HTTP server. See the documentation of tf.io.http for more details including specifying request parameters and implementation of the server.
const model = tf.sequential( {layers: [tf.layers.dense({units: 1, inputShape: [3]})]}); const saveResults = await model.save('http://my-server/model/upload');
Parameter handlerOrURL
An instance of IOHandler or a URL-like, scheme-based string shortcut for IOHandler.
Parameter config
Options for saving the model.
Returns
A Promise of SaveResult, which summarizes the result of the saving, such as byte sizes of the saved artifacts for the model's topology and weight values.
{heading: 'Models', subheading: 'Classes', ignoreCI: true}

method setUserDefinedMetadata

setUserDefinedMetadata: (userDefinedMetadata: {}) => void;

Set user-defined metadata.
The set metadata will be serialized together with the topology and weights of the model during save() calls.
Parameter setUserDefinedMetadata

method standardizeUserData

protected standardizeUserData: (
    x: any,
    y: any,
    sampleWeight?: any,
    classWeight?: ClassWeight | ClassWeight[] | ClassWeightMap,
    checkBatchAxis?: boolean,
    batchSize?: number
) => Promise<[Tensor[], Tensor[], Tensor[]]>;

method standardizeUserDataXY

protected standardizeUserDataXY: (
    x: any,
    y: any,
    checkBatchAxis?: boolean,
    batchSize?: number
) => [Tensor[], Tensor[]];

method summary

summary: (
    lineLength?: number,
    positions?: number[],
    printFn?: (message?: any, ...optionalParams: any[]) => void
) => void;

Print a text summary of the model's layers.
The summary includes - Name and type of all layers that comprise the model. - Output shape(s) of the layers - Number of weight parameters of each layer - If the model has non-sequential-like topology, the inputs each layer receives - The total number of trainable and non-trainable parameters of the model.
const input1 = tf.input({shape: [10]}); const input2 = tf.input({shape: [20]}); const dense1 = tf.layers.dense({units: 4}).apply(input1); const dense2 = tf.layers.dense({units: 8}).apply(input2); const concat = tf.layers.concatenate().apply([dense1, dense2]); const output = tf.layers.dense({units: 3, activation: 'softmax'}).apply(concat); const model = tf.model({inputs: [input1, input2], outputs: output}); model.summary();
Parameter lineLength
Custom line length, in number of characters.
Parameter positions
Custom widths of each of the columns, as either fractions of lineLength (e.g., [0.5, 0.75, 1]) or absolute number of characters (e.g., [30, 50, 65]). Each number corresponds to right-most (i.e., ending) position of a column.
Parameter printFn
Custom print function. Can be used to replace the default console.log. For example, you can use x => {} to mute the printed messages in the console.
{heading: 'Models', subheading: 'Classes'}

method trainOnBatch

trainOnBatch: (x: any, y: any) => Promise<number | number[]>;

Runs a single gradient update on a single batch of data.
This method differs from fit() and fitDataset() in the following regards: - It operates on exactly one batch of data. - It returns only the loss and metric values, instead of returning the batch-by-batch loss and metric values. - It doesn't support fine-grained options such as verbosity and callbacks.
Parameter x
Input data. It could be one of the following: - A tf.Tensor, or an Array of tf.Tensors (in case the model has multiple inputs). - An Object mapping input names to corresponding tf.Tensor (if the model has named inputs).
Parameter y
Target data. It could be either a tf.Tensor or multiple tf.Tensors. It should be consistent with x.
Returns
Training loss or losses (in case the model has multiple outputs), along with metrics (if any), as numbers.
{heading: 'Models', subheading: 'Classes'}

class LayerVariable

class LayerVariable {}

A tf.layers.LayerVariable is similar to a tf.Tensor in that it has a dtype and shape, but its value is mutable. The value is itself represented as atf.Tensor, and can be read with the read() method and updated with the write() method.

constructor

constructor(
    val: Tensor,
    dtype?: DataType,
    name?: string,
    trainable?: boolean,
    constraint?: Constraint
);

Construct Variable from a tf.Tensor.
If not explicitly named, the Variable will be given a name with the prefix 'Variable'. Variable names are unique. In the case of name collision, suffixies '_' will be added to the name.
Parameter val
Initial value of the Variable.
Parameter name
Name of the variable. If null or undefined is provided, it will default a name with the prefix 'Variable'.
Parameter constraint
Optional, projection function to be applied to the variable after optimize updates
Throws
ValueError if name is null or undefined.

property constraint

readonly constraint: Constraint;

property dtype

readonly dtype: DataType;

property id

readonly id: number;

property name

readonly name: string;

property originalName

readonly originalName: string;

property shape

readonly shape: Shape;

property trainable

trainable: boolean;

property val

protected readonly val: tfc.Variable;

method assertNotDisposed

protected assertNotDisposed: () => void;

method dispose

dispose: () => void;

Dispose this LayersVariable instance from memory.

method read

read: () => Tensor;

Get a snapshot of the Variable's value.
The returned value is a snapshot of the Variable's value at the time of the invocation. Future mutations in the value of the tensor will only be reflected by future calls to this method.

method write

write: (newVal: Tensor) => this;

Update the value of the Variable.
Parameter newVal
: The new value to update to. Must be consistent with the dtype and shape of the Variable. This Variable.

class RNN

class RNN extends Layer {}

constructor

constructor(args: RNNLayerArgs);

property cell

readonly cell: RNNCell;

property className

static className: string;

property goBackwards

readonly goBackwards: boolean;

property keptStates

protected keptStates: Tensor[][];

property nonTrainableWeights

readonly nonTrainableWeights: LayerVariable[];

property returnSequences

readonly returnSequences: boolean;

property returnState

readonly returnState: boolean;

property states

states: Tensor[];

Get the current state tensors of the RNN.
If the state hasn't been set, return an array of nulls of the correct length.

property states_

protected states_: Tensor[];

property stateSpec

stateSpec: InputSpec[];

property trainableWeights

readonly trainableWeights: LayerVariable[];

property unroll

readonly unroll: boolean;

method apply

apply: (
    inputs: Tensor | Tensor[] | SymbolicTensor | SymbolicTensor[],
    kwargs?: Kwargs
) => Tensor | Tensor[] | SymbolicTensor | SymbolicTensor[];

method build

build: (inputShape: Shape | Shape[]) => void;

method call

call: (inputs: Tensor | Tensor[], kwargs: Kwargs) => Tensor | Tensor[];

method computeMask

computeMask: (
    inputs: Tensor | Tensor[],
    mask?: Tensor | Tensor[]
) => Tensor | Tensor[];

method computeOutputShape

computeOutputShape: (inputShape: Shape | Shape[]) => Shape | Shape[];

method fromConfig

static fromConfig: <T extends serialization.Serializable>(
    cls: serialization.SerializableConstructor<T>,
    config: serialization.ConfigDict,
    customObjects?: serialization.ConfigDict
) => T;

method getConfig

getConfig: () => serialization.ConfigDict;

method getInitialState

getInitialState: (inputs: Tensor) => Tensor[];

method getStates

getStates: () => Tensor[];

method resetStates

resetStates: (states?: Tensor | Tensor[], training?: boolean) => void;

Reset the state tensors of the RNN.
If the states argument is undefined or null, will set the state tensor(s) of the RNN to all-zero tensors of the appropriate shape(s).
If states is provided, will set the state tensors of the RNN to its value.
Parameter states
Optional externally-provided initial states.
Parameter training
Whether this call is done during training. For stateful RNNs, this affects whether the old states are kept or discarded. In particular, if training is true, the old states will be kept so that subsequent backpropgataion through time (BPTT) may work properly. Else, the old states will be discarded.

method setFastWeightInitDuringBuild

setFastWeightInitDuringBuild: (value: boolean) => void;

method setStates

setStates: (states: Tensor[]) => void;

class Sequential

class Sequential extends LayersModel {}

A model with a stack of layers, feeding linearly from one to the next.

tf.sequential is a factory function that creates an instance of tf.Sequential.

 // Define a model for linear regression.
 const model = tf.sequential();
 model.add(tf.layers.dense({units: 1, inputShape: [1]}));

 // Prepare the model for training: Specify the loss and the optimizer.
 model.compile({loss: 'meanSquaredError', optimizer: 'sgd'});

 // Generate some synthetic data for training.
 const xs = tf.tensor2d([1, 2, 3, 4], [4, 1]);
 const ys = tf.tensor2d([1, 3, 5, 7], [4, 1]);

 // Train the model using the data then do inference on a data point the
 // model hasn't seen:
 await model.fit(xs, ys);
 model.predict(tf.tensor2d([5], [1, 1])).print();

{heading: 'Models', subheading: 'Classes'}

constructor

constructor(args?: SequentialArgs);

property className

static className: string;

property optimizer

optimizer: Optimizer;

property stopTraining

stopTraining: boolean;

method add

add: (layer: Layer) => void;

Adds a layer instance on top of the layer stack.

 const model = tf.sequential();
 model.add(tf.layers.dense({units: 8, inputShape: [1]}));
 model.add(tf.layers.dense({units: 4, activation: 'relu6'}));
 model.add(tf.layers.dense({units: 1, activation: 'relu6'}));
 // Note that the untrained model is random at this point.
 model.predict(tf.randomNormal([10, 1])).print();

Parameter layer

Layer instance.

ValueError In case the layer argument does not know its input shape. ValueError In case the layer argument has multiple output tensors, or is already connected somewhere else (forbidden in Sequential models).

{heading: 'Models', subheading: 'Classes'}

method build

build: (inputShape?: Shape | Shape[]) => void;

method call

call: (inputs: Tensor | Tensor[], kwargs: Kwargs) => Tensor | Tensor[];

method compile

compile: (args: ModelCompileArgs) => void;

See LayersModel.compile.
Parameter args

method countParams

countParams: () => number;

method evaluate

evaluate: (
    x: Tensor | Tensor[],
    y: Tensor | Tensor[],
    args?: ModelEvaluateArgs
) => Scalar | Scalar[];

Returns the loss value & metrics values for the model in test mode.
Loss and metrics are specified during compile(), which needs to happen before calls to evaluate().
Computation is done in batches.
const model = tf.sequential({ layers: [tf.layers.dense({units: 1, inputShape: [10]})] }); model.compile({optimizer: 'sgd', loss: 'meanSquaredError'}); const result = model.evaluate(tf.ones([8, 10]), tf.ones([8, 1]), { batchSize: 4, }); result.print();
Parameter x
tf.Tensor of test data, or an Array of tf.Tensors if the model has multiple inputs.
Parameter y
tf.Tensor of target data, or an Array of tf.Tensors if the model has multiple outputs.
Parameter args
A ModelEvaluateConfig, containing optional fields.
Scalar test loss (if the model has a single output and no metrics) or Array of Scalars (if the model has multiple outputs and/or metrics). The attribute model.metricsNames will give you the display labels for the scalar outputs.
{heading: 'Models', subheading: 'Classes'}

method evaluateDataset

evaluateDataset: (
    dataset: Dataset<{}>,
    args: ModelEvaluateDatasetArgs
) => Promise<Scalar | Scalar[]>;

Evaluate model using a dataset object.
Note: Unlike evaluate(), this method is asynchronous (async).
Parameter dataset
A dataset object. Its iterator() method is expected to generate a dataset iterator object, the next() method of which is expected to produce data batches for evaluation. The return value of the next() call ought to contain a boolean done field and a value field. The value field is expected to be an array of two tf.Tensors or an array of two nested tf.Tensor structures. The former case is for models with exactly one input and one output (e.g. a sequential model). The latter case is for models with multiple inputs and/or multiple outputs. Of the two items in the array, the first is the input feature(s) and the second is the output target(s).
Parameter args
A configuration object for the dataset-based evaluation.
Returns
Loss and metric values as an Array of Scalar objects.
{heading: 'Models', subheading: 'Classes'}

method fit

fit: (x: any, y: any, args?: ModelFitArgs) => Promise<History>;

Trains the model for a fixed number of epochs (iterations on a dataset).
const model = tf.sequential({ layers: [tf.layers.dense({units: 1, inputShape: [10]})] }); model.compile({optimizer: 'sgd', loss: 'meanSquaredError'}); const history = await model.fit(tf.ones([8, 10]), tf.ones([8, 1]), { batchSize: 4, epochs: 3 }); console.log(history.history.loss[0]);
Parameter x
tf.Tensor of training data, or an array of tf.Tensors if the model has multiple inputs. If all inputs in the model are named, you can also pass a dictionary mapping input names to tf.Tensors.
Parameter y
tf.Tensor of target (label) data, or an array of tf.Tensors if the model has multiple outputs. If all outputs in the model are named, you can also pass a dictionary mapping output names to tf.Tensors.
Parameter args
A ModelFitConfig, containing optional fields.
A History instance. Its history attribute contains all information collected during training.
ValueError In case of mismatch between the provided input data and what the model expects.
{heading: 'Models', subheading: 'Classes'}

method fitDataset

fitDataset: <T>(
    dataset: Dataset<T>,
    args: ModelFitDatasetArgs<T>
) => Promise<History>;

Trains the model using a dataset object.
const xArray = [ [1, 1, 1, 1, 1, 1, 1, 1, 1], [1, 1, 1, 1, 1, 1, 1, 1, 1], [1, 1, 1, 1, 1, 1, 1, 1, 1], [1, 1, 1, 1, 1, 1, 1, 1, 1], ]; const yArray = [1, 1, 1, 1]; // Create a dataset from the JavaScript array. const xDataset = tf.data.array(xArray); const yDataset = tf.data.array(yArray); // Zip combines the `x` and `y` Datasets into a single Dataset, the // iterator of which will return an object containing of two tensors, // corresponding to `x` and `y`. The call to `batch(4)` will bundle // four such samples into a single object, with the same keys now pointing // to tensors that hold 4 examples, organized along the batch dimension. // The call to `shuffle(4)` causes each iteration through the dataset to // happen in a different order. The size of the shuffle window is 4. const xyDataset = tf.data.zip({xs: xDataset, ys: yDataset}) .batch(4) .shuffle(4); const model = tf.sequential({ layers: [tf.layers.dense({units: 1, inputShape: [9]})] }); model.compile({optimizer: 'sgd', loss: 'meanSquaredError'}); const history = await model.fitDataset(xyDataset, { epochs: 4, callbacks: {onEpochEnd: (epoch, logs) => console.log(logs.loss)} });
Parameter dataset
A dataset object. Its iterator() method is expected to generate a dataset iterator object, the next() method of which is expected to produce data batches for evaluation. The return value of the next() call ought to contain a boolean done field and a value field.
The value field is expected to be an object of with fields xs and ys, which point to the feature tensor and the target tensor, respectively. This case is for models with exactly one input and one output (e.g. a sequential model). For example: ```js {value: {xs: xsTensor, ys: ysTensor}, done: false} ```
If the model has multiple inputs, the xs field of value should be an object mapping input names to their respective feature tensors. For example: ```js { value: { xs: { input_1: xsTensor1, input_2: xsTensor2 }, ys: ysTensor }, done: false } ``` If the model has multiple outputs, the ys field of value should be an object mapping output names to their respective target tensors. For example: ```js { value: { xs: xsTensor, ys: { output_1: ysTensor1, output_2: ysTensor2 }, }, done: false } ```
Parameter args
A ModelFitDatasetArgs, containing optional fields.
A History instance. Its history attribute contains all information collected during training.
{heading: 'Models', subheading: 'Classes', ignoreCI: true}

method fromConfig

static fromConfig: <T extends serialization.Serializable>(
    cls: serialization.SerializableConstructor<T>,
    config: serialization.ConfigDict,
    customObjects?: serialization.ConfigDict,
    fastWeightInit?: boolean
) => T;

method getConfig

getConfig: () => any;

method pop

pop: () => void;

Removes the last layer in the model.
TypeError if there are no layers in the model.

method predict

predict: (x: Tensor | Tensor[], args?: ModelPredictArgs) => Tensor | Tensor[];

Generates output predictions for the input samples.
Computation is done in batches.
Note: the "step" mode of predict() is currently not supported. This is because the TensorFlow.js core backend is imperative only.
const model = tf.sequential({ layers: [tf.layers.dense({units: 1, inputShape: [10]})] }); model.predict(tf.ones([2, 10])).print();
Parameter x
The input data, as a Tensor, or an Array of tf.Tensors if the model has multiple inputs.
Parameter conifg
A ModelPredictConfig object containing optional fields.
tf.Tensor(s) of predictions.
ValueError In case of mismatch between the provided input data and the model's expectations, or in case a stateful model receives a number of samples that is not a multiple of the batch size.
{heading: 'Models', subheading: 'Classes'}

method predictOnBatch

predictOnBatch: (x: Tensor) => Tensor | Tensor[];

Returns predictions for a single batch of samples.
Parameter x
: Input samples, as a Tensor, or list of Tensors (if the model has multiple inputs). Tensor(s) of predictions

method setWeights

setWeights: (weights: Tensor[]) => void;

Sets the weights of the model.
Parameter weights
Should be a list of Tensors with shapes and types matching the output of model.getWeights().

method summary

summary: (
    lineLength?: number,
    positions?: number[],
    printFn?: (message?: any, ...optionalParams: any[]) => void
) => void;

Print a text summary of the Sequential model's layers.
The summary includes - Name and type of all layers that comprise the model. - Output shape(s) of the layers - Number of weight parameters of each layer - The total number of trainable and non-trainable parameters of the model.
const model = tf.sequential(); model.add( tf.layers.dense({units: 100, inputShape: [10], activation: 'relu'})); model.add(tf.layers.dense({units: 1, activation: 'sigmoid'})); model.summary();
Parameter lineLength
Custom line length, in number of characters.
Parameter positions
Custom widths of each of the columns, as either fractions of lineLength (e.g., [0.5, 0.75, 1]) or absolute number of characters (e.g., [30, 50, 65]). Each number corresponds to right-most (i.e., ending) position of a column.
Parameter printFn
Custom print function. Can be used to replace the default console.log. For example, you can use x => {} to mute the printed messages in the console.
{heading: 'Models', subheading: 'Classes'}

method trainOnBatch

trainOnBatch: (x: any, y: any) => Promise<number | number[]>;

Runs a single gradient update on a single batch of data.
This method differs from fit() and fitDataset() in the following regards: - It operates on exactly one batch of data. - It returns only the loss and metric values, instead of returning the batch-by-batch loss and metric values. - It doesn't support fine-grained options such as verbosity and callbacks.
Parameter x
Input data. It could be one of the following: - A tf.Tensor, or an Array of tf.Tensors (in case the model has multiple inputs). - An Object mapping input names to corresponding tf.Tensor (if the model has named inputs).
Parameter y
Target data. It could be either a tf.Tensor or multiple tf.Tensors. It should be consistent with x.
Returns
Training loss or losses (in case the model has multiple outputs), along with metrics (if any), as numbers.
{heading: 'Models', subheading: 'Classes'}

class SymbolicTensor

class SymbolicTensor {}

tf.SymbolicTensor is a placeholder for a Tensor without any concrete value.
They are most often encountered when building a graph of Layers for a tf.LayersModel and the input data's shape, but not values are known.
{heading: 'Models', 'subheading': 'Classes'}

constructor

constructor(
    dtype: DataType,
    shape: Shape,
    sourceLayer: Layer,
    inputs: SymbolicTensor[],
    callArgs: Kwargs,
    name?: string,
    outputTensorIndex?: number
);

Parameter dtype
Parameter shape
Parameter sourceLayer
The Layer that produced this symbolic tensor.
Parameter inputs
The inputs passed to sourceLayer's __call__() method.
Parameter nodeIndex
Parameter tensorIndex
Parameter callArgs
The keyword arguments passed to the __call__() method.
Parameter name
Parameter outputTensorIndex
The index of this tensor in the list of outputs returned by apply().

property callArgs

readonly callArgs: Kwargs;

property dtype

readonly dtype: DataType;

property id

readonly id: number;

property inputs

readonly inputs: SymbolicTensor[];

property name

readonly name: string;

property nodeIndex

nodeIndex: number;

Replacement for _keras_history.

property originalName

readonly originalName?: string;

property outputTensorIndex

readonly outputTensorIndex?: number;

property rank

readonly rank: number;

Rank/dimensionality of the tensor.

property shape

readonly shape: Shape;

property sourceLayer

sourceLayer: Layer;

property tensorIndex

tensorIndex: number;

Replacement for _keras_history.

Interfaces

interface CustomCallbackArgs

interface CustomCallbackArgs {}

property nextFrameFunc

nextFrameFunc?: Function;

property nowFunc

nowFunc?: Function;

property onBatchBegin

onBatchBegin?: (batch: number, logs?: Logs) => void | Promise<void>;

property onBatchEnd

onBatchEnd?: (batch: number, logs?: Logs) => void | Promise<void>;

property onEpochBegin

onEpochBegin?: (epoch: number, logs?: Logs) => void | Promise<void>;

property onEpochEnd

onEpochEnd?: (epoch: number, logs?: Logs) => void | Promise<void>;

property onTrainBegin

onTrainBegin?: (logs?: Logs) => void | Promise<void>;

property onTrainEnd

onTrainEnd?: (logs?: Logs) => void | Promise<void>;

property onYield

onYield?: (epoch: number, batch: number, logs: Logs) => void | Promise<void>;

interface EarlyStoppingCallbackArgs

interface EarlyStoppingCallbackArgs {}

property baseline

baseline?: number;

Baseline value of the monitored quantity.
If specified, training will be stopped if the model doesn't show improvement over the baseline.

property minDelta

minDelta?: number;

Minimum change in the monitored quantity to qualify as improvement, i.e., an absolute change of less than minDelta will count as no improvement.
Defaults to 0.

property mode

mode?: 'auto' | 'min' | 'max';

Mode: one of 'min', 'max', and 'auto'. - In 'min' mode, training will be stopped when the quantity monitored has stopped decreasing. - In 'max' mode, training will be stopped when the quantity monitored has stopped increasing. - In 'auto' mode, the direction is inferred automatically from the name of the monitored quantity.
Defaults to 'auto'.

property monitor

monitor?: string;

Quantity to be monitored.
Defaults to 'val_loss'.

property patience

patience?: number;

Number of epochs with no improvement after which training will be stopped.
Defaults to 0.

property restoreBestWeights

restoreBestWeights?: boolean;

Whether to restore model weights from the epoch with the best value of the monitored quantity. If False, the model weights obtained at the last step of training are used.
**True is not supported yet.**

property verbose

verbose?: number;

Verbosity mode.

interface GRUCellLayerArgs

interface GRUCellLayerArgs extends SimpleRNNCellLayerArgs {}

property implementation

implementation?: number;

Implementation mode, either 1 or 2.
Mode 1 will structure its operations as a larger number of smaller dot products and additions.
Mode 2 will batch them into fewer, larger operations. These modes will have different performance profiles on different hardware and for different applications.
Note: For superior performance, TensorFlow.js always uses implementation 2, regardless of the actual value of this configuration field.

property recurrentActivation

recurrentActivation?: ActivationIdentifier;

Activation function to use for the recurrent step.
Defaults to hard sigmoid (hardSigmoid).
If null, no activation is applied.

property resetAfter

resetAfter?: boolean;

GRU convention (whether to apply reset gate after or before matrix multiplication). false = "before", true = "after" (only false is supported).

interface GRULayerArgs

interface GRULayerArgs extends SimpleRNNLayerArgs {}

property implementation

implementation?: number;

Implementation mode, either 1 or 2.
Mode 1 will structure its operations as a larger number of smaller dot products and additions.
Mode 2 will batch them into fewer, larger operations. These modes will have different performance profiles on different hardware and for different applications.
Note: For superior performance, TensorFlow.js always uses implementation 2, regardless of the actual value of this configuration field.

property recurrentActivation

recurrentActivation?: ActivationIdentifier;

Activation function to use for the recurrent step.
Defaults to hard sigmoid (hardSigmoid).
If null, no activation is applied.

interface LSTMCellLayerArgs

interface LSTMCellLayerArgs extends SimpleRNNCellLayerArgs {}

property implementation

implementation?: number;

Implementation mode, either 1 or 2.
Mode 1 will structure its operations as a larger number of smaller dot products and additions.
Mode 2 will batch them into fewer, larger operations. These modes will have different performance profiles on different hardware and for different applications.
Note: For superior performance, TensorFlow.js always uses implementation 2, regardless of the actual value of this configuration field.

property recurrentActivation

recurrentActivation?: ActivationIdentifier;

Activation function to use for the recurrent step.
Defaults to hard sigmoid (hardSigmoid).
If null, no activation is applied.

property unitForgetBias

unitForgetBias?: boolean;

If true, add 1 to the bias of the forget gate at initialization. Setting it to true will also force biasInitializer = 'zeros'. This is recommended in [Jozefowicz et al.](http://www.jmlr.org/proceedings/papers/v37/jozefowicz15.pdf)

interface LSTMLayerArgs

interface LSTMLayerArgs extends SimpleRNNLayerArgs {}

property implementation

implementation?: number;

Implementation mode, either 1 or 2. Mode 1 will structure its operations as a larger number of smaller dot products and additions, whereas mode 2 will batch them into fewer, larger operations. These modes will have different performance profiles on different hardware and for different applications.
Note: For superior performance, TensorFlow.js always uses implementation 2, regardless of the actual value of this config field.

property recurrentActivation

recurrentActivation?: ActivationIdentifier;

Activation function to use for the recurrent step.
Defaults to hard sigmoid (hardSigmoid).
If null, no activation is applied.

property unitForgetBias

unitForgetBias?: boolean;

If true, add 1 to the bias of the forget gate at initialization. Setting it to true will also force biasInitializer = 'zeros'. This is recommended in [Jozefowicz et al.](http://www.jmlr.org/proceedings/papers/v37/jozefowicz15.pdf)

interface ModelAndWeightsConfig

interface ModelAndWeightsConfig {}

Options for loading a saved mode in TensorFlow.js format.

property modelTopology

modelTopology: PyJsonDict;

A JSON object or JSON string containing the model config.
This can be either of the following two formats: - A model archiecture-only config, i.e., a format consistent with the return value ofkeras.Model.to_json(). - A full model config, containing not only model architecture, but also training options and state, i.e., a format consistent with the return value of keras.models.save_model().

property pathPrefix

pathPrefix?: string;

Path to prepend to the paths in weightManifest before fetching.
The path may optionally end in a slash ('/').

property weightsManifest

weightsManifest?: io.WeightsManifestConfig;

A weights manifest in TensorFlow.js format.

interface ModelCompileArgs

interface ModelCompileArgs {}

Configuration for calls to LayersModel.compile().

property loss

loss:
    | string
    | string[]
    | {
          [outputName: string]: string;
      }
    | LossOrMetricFn
    | LossOrMetricFn[]
    | {
          [outputName: string]: LossOrMetricFn;
      };

Object function(s) or name(s) of object function(s). If the model has multiple outputs, you can use a different loss on each output by passing a dictionary or an Array of losses. The loss value that will be minimized by the model will then be the sum of all individual losses.

property metrics

metrics?:
    | string
    | LossOrMetricFn
    | Array<string | LossOrMetricFn>
    | {
          [outputName: string]: string | LossOrMetricFn;
      };

List of metrics to be evaluated by the model during training and testing. Typically you will use metrics=['accuracy']. To specify different metrics for different outputs of a multi-output model, you could also pass a dictionary.

property optimizer

optimizer: string | Optimizer;

An instance of tf.train.Optimizer or a string name for an Optimizer.

interface ModelEvaluateArgs

interface ModelEvaluateArgs {}

property batchSize

batchSize?: number;

Batch size (Integer). If unspecified, it will default to 32.

property sampleWeight

sampleWeight?: Tensor;

Tensor of weights to weight the contribution of different samples to the loss and metrics.

property steps

steps?: number;

integer: total number of steps (batches of samples) before declaring the evaluation round finished. Ignored with the default value of undefined.

property verbose

verbose?: ModelLoggingVerbosity;

Verbosity mode.

interface ModelFitArgs

interface ModelFitArgs {}

Interface configuration model training based on data as tf.Tensors.

property batchSize

batchSize?: number;

Number of samples per gradient update. If unspecified, it will default to 32.

property callbacks

callbacks?: BaseCallback[] | CustomCallbackArgs | CustomCallbackArgs[];

List of callbacks to be called during training. Can have one or more of the following callbacks: - onTrainBegin(logs): called when training starts. - onTrainEnd(logs): called when training ends. - onEpochBegin(epoch, logs): called at the start of every epoch. - onEpochEnd(epoch, logs): called at the end of every epoch. - onBatchBegin(batch, logs): called at the start of every batch. - onBatchEnd(batch, logs): called at the end of every batch. - onYield(epoch, batch, logs): called every yieldEvery milliseconds with the current epoch, batch and logs. The logs are the same as in onBatchEnd(). Note that onYield can skip batches or epochs. See also docs for yieldEvery below.

property classWeight

classWeight?: ClassWeight | ClassWeight[] | ClassWeightMap;

Optional object mapping class indices (integers) to a weight (float) to apply to the model's loss for the samples from this class during training. This can be useful to tell the model to "pay more attention" to samples from an under-represented class.
If the model has multiple outputs, a class weight can be specified for each of the outputs by setting this field an array of weight object or an object that maps model output names (e.g., model.outputNames[0]) to weight objects.

property epochs

epochs?: number;

Integer number of times to iterate over the training data arrays.

property initialEpoch

initialEpoch?: number;

Epoch at which to start training (useful for resuming a previous training run). When this is used, epochs is the index of the "final epoch". The model is not trained for a number of iterations given by epochs, but merely until the epoch of index epochs is reached.

property sampleWeight

sampleWeight?: Tensor;

Optional array of the same length as x, containing weights to apply to the model's loss for each sample. In the case of temporal data, you can pass a 2D array with shape (samples, sequenceLength), to apply a different weight to every timestep of every sample. In this case you should make sure to specify sampleWeightMode="temporal" in compile().

property shuffle

shuffle?: boolean;

Whether to shuffle the training data before each epoch. Has no effect when stepsPerEpoch is not null.

property stepsPerEpoch

stepsPerEpoch?: number;

Total number of steps (batches of samples) before declaring one epoch finished and starting the next epoch. When training with Input Tensors such as TensorFlow data tensors, the default null is equal to the number of unique samples in your dataset divided by the batch size, or 1 if that cannot be determined.

property validationData

validationData?:
    | [Tensor | Tensor[], Tensor | Tensor[]]
    | [Tensor | Tensor[], Tensor | Tensor[], Tensor | Tensor[]];

Data on which to evaluate the loss and any model metrics at the end of each epoch. The model will not be trained on this data. This could be a tuple [xVal, yVal] or a tuple [xVal, yVal, valSampleWeights]. The model will not be trained on this data. validationData will override validationSplit.

property validationSplit

validationSplit?: number;

Float between 0 and 1: fraction of the training data to be used as validation data. The model will set apart this fraction of the training data, will not train on it, and will evaluate the loss and any model metrics on this data at the end of each epoch. The validation data is selected from the last samples in the x and y data provided, before shuffling.

property validationSteps

validationSteps?: number;

Only relevant if stepsPerEpoch is specified. Total number of steps (batches of samples) to validate before stopping.

property verbose

verbose?: ModelLoggingVerbosity | 2;

Verbosity level.
Expected to be 0, 1, or 2. Default: 1.
0 - No printed message during fit() call. 1 - In Node.js (tfjs-node), prints the progress bar, together with real-time updates of loss and metric values and training speed. In the browser: no action. This is the default. 2 - Not implemented yet.

property yieldEvery

yieldEvery?: YieldEveryOptions;

Configures the frequency of yielding the main thread to other tasks.
In the browser environment, yielding the main thread can improve the responsiveness of the page during training. In the Node.js environment, it can ensure tasks queued in the event loop can be handled in a timely manner.
The value can be one of the following: - 'auto': The yielding happens at a certain frame rate (currently set at 125ms). This is the default. - 'batch': yield every batch. - 'epoch': yield every epoch. - any number: yield every number milliseconds. - 'never': never yield. (yielding can still happen through `await nextFrame()` calls in custom callbacks.)

interface ModelFitDatasetArgs

interface ModelFitDatasetArgs<T> {}

Interface for configuring model training based on a dataset object.

property batchesPerEpoch

batchesPerEpoch?: number;

(Optional) Total number of steps (batches of samples) before declaring one epoch finished and starting the next epoch. It should typically be equal to the number of samples of your dataset divided by the batch size, so that fitDataset() call can utilize the entire dataset. If it is not provided, use done return value in iterator.next() as signal to finish an epoch.

property callbacks

callbacks?: BaseCallback[] | CustomCallbackArgs | CustomCallbackArgs[];

List of callbacks to be called during training. Can have one or more of the following callbacks: - onTrainBegin(logs): called when training starts. - onTrainEnd(logs): called when training ends. - onEpochBegin(epoch, logs): called at the start of every epoch. - onEpochEnd(epoch, logs): called at the end of every epoch. - onBatchBegin(batch, logs): called at the start of every batch. - onBatchEnd(batch, logs): called at the end of every batch. - onYield(epoch, batch, logs): called every yieldEvery milliseconds with the current epoch, batch and logs. The logs are the same as in onBatchEnd(). Note that onYield can skip batches or epochs. See also docs for yieldEvery below.

property classWeight

classWeight?: ClassWeight | ClassWeight[] | ClassWeightMap;

Optional object mapping class indices (integers) to a weight (float) to apply to the model's loss for the samples from this class during training. This can be useful to tell the model to "pay more attention" to samples from an under-represented class.
If the model has multiple outputs, a class weight can be specified for each of the outputs by setting this field an array of weight object or an object that maps model output names (e.g., model.outputNames[0]) to weight objects.

property epochs

epochs: number;

Integer number of times to iterate over the training dataset.

property initialEpoch

initialEpoch?: number;

Epoch at which to start training (useful for resuming a previous training run). When this is used, epochs is the index of the "final epoch". The model is not trained for a number of iterations given by epochs, but merely until the epoch of index epochs is reached.

property validationBatches

validationBatches?: number;

(Optional) Only relevant if validationData is specified and is a dataset object.
Total number of batches of samples to draw from validationData for validation purpose before stopping at the end of every epoch. If not specified, evaluateDataset will use iterator.next().done as signal to stop validation.

property validationBatchSize

validationBatchSize?: number;

Optional batch size for validation.
Used only if validationData is an array of tf.Tensor objects, i.e., not a dataset object.
If not specified, its value defaults to 32.

property validationData

validationData?:
    | [TensorOrArrayOrMap, TensorOrArrayOrMap]
    | [TensorOrArrayOrMap, TensorOrArrayOrMap, TensorOrArrayOrMap]
    | Dataset<T>;

Data on which to evaluate the loss and any model metrics at the end of each epoch. The model will not be trained on this data. This could be any of the following:
- An array [xVal, yVal], where the two values may be tf.Tensor, an array of Tensors, or a map of string to Tensor. - Similarly, an array [xVal, yVal, valSampleWeights] (not implemented yet). - a Dataset object with elements of the form {xs: xVal, ys: yVal}, where xs and ys are the feature and label tensors, respectively.
If validationData is an Array of Tensor objects, each tf.Tensor will be sliced into batches during validation, using the parameter validationBatchSize (which defaults to 32). The entirety of the tf.Tensor objects will be used in the validation.
If validationData is a dataset object, and the validationBatches parameter is specified, the validation will use validationBatches batches drawn from the dataset object. If validationBatches parameter is not specified, the validation will stop when the dataset is exhausted.
The model will not be trained on this data.

property verbose

verbose?: ModelLoggingVerbosity;

Verbosity level.
Expected to be 0, 1, or 2. Default: 1.
0 - No printed message during fit() call. 1 - In Node.js (tfjs-node), prints the progress bar, together with real-time updates of loss and metric values and training speed. In the browser: no action. This is the default. 2 - Not implemented yet.

property yieldEvery

yieldEvery?: YieldEveryOptions;

Configures the frequency of yielding the main thread to other tasks.
In the browser environment, yielding the main thread can improve the responsiveness of the page during training. In the Node.js environment, it can ensure tasks queued in the event loop can be handled in a timely manner.
The value can be one of the following: - 'auto': The yielding happens at a certain frame rate (currently set at 125ms). This is the default. - 'batch': yield every batch. - 'epoch': yield every epoch. - a number: Will yield every number milliseconds. - 'never': never yield. (But yielding can still happen through `await nextFrame()` calls in custom callbacks.)

interface RNNLayerArgs

interface RNNLayerArgs extends BaseRNNLayerArgs {}

RNNLayerConfig is identical to BaseRNNLayerConfig, except it makes the cell property required. This interface is to be used with constructors of concrete RNN layer subtypes.

property cell

cell: RNNCell | RNNCell[];

interface SequentialArgs

interface SequentialArgs {}

Configuration for a Sequential model.

property layers

layers?: Layer[];

Stack of layers for the model.

property name

name?: string;

The name of this model.

interface SimpleRNNCellLayerArgs

interface SimpleRNNCellLayerArgs extends LayerArgs {}

property activation

activation?: ActivationIdentifier;

Activation function to use. Default: hyperbolic tangent ('tanh'). If you pass null, 'linear' activation will be applied.

property biasConstraint

biasConstraint?: ConstraintIdentifier | Constraint;

Constraint function applied to the bias vector.

property biasInitializer

biasInitializer?: InitializerIdentifier | Initializer;

Initializer for the bias vector.

property biasRegularizer

biasRegularizer?: RegularizerIdentifier | Regularizer;

Regularizer function applied to the bias vector.

property dropout

dropout?: number;

Float number between 0 and 1. Fraction of the units to drop for the linear transformation of the inputs.

property dropoutFunc

dropoutFunc?: Function;

This is added for test DI purpose.

property kernelConstraint

kernelConstraint?: ConstraintIdentifier | Constraint;

Constraint function applied to the kernel weights matrix.

property kernelInitializer

kernelInitializer?: InitializerIdentifier | Initializer;

Initializer for the kernel weights matrix, used for the linear transformation of the inputs.

property kernelRegularizer

kernelRegularizer?: RegularizerIdentifier | Regularizer;

Regularizer function applied to the kernel weights matrix.

property recurrentConstraint

recurrentConstraint?: ConstraintIdentifier | Constraint;

Constraint function applied to the recurrentKernel weights matrix.

property recurrentDropout

recurrentDropout?: number;

Float number between 0 and 1. Fraction of the units to drop for the linear transformation of the recurrent state.

property recurrentInitializer

recurrentInitializer?: InitializerIdentifier | Initializer;

Initializer for the recurrentKernel weights matrix, used for linear transformation of the recurrent state.

property recurrentRegularizer

recurrentRegularizer?: RegularizerIdentifier | Regularizer;

Regularizer function applied to the recurrent_kernel weights matrix.

property units

units: number;

units: Positive integer, dimensionality of the output space.

property useBias

useBias?: boolean;

Whether the layer uses a bias vector.

interface SimpleRNNLayerArgs

interface SimpleRNNLayerArgs extends BaseRNNLayerArgs {}

property activation

activation?: ActivationIdentifier;

Activation function to use.
Defaults to hyperbolic tangent (tanh)
If you pass null, no activation will be applied.

property biasConstraint

biasConstraint?: ConstraintIdentifier | Constraint;

Constraint function applied to the bias vector.

property biasInitializer

biasInitializer?: InitializerIdentifier | Initializer;

Initializer for the bias vector.

property biasRegularizer

biasRegularizer?: RegularizerIdentifier | Regularizer;

Regularizer function applied to the bias vector.

property dropout

dropout?: number;

Number between 0 and 1. Fraction of the units to drop for the linear transformation of the inputs.

property dropoutFunc

dropoutFunc?: Function;

This is added for test DI purpose.

property kernelConstraint

kernelConstraint?: ConstraintIdentifier | Constraint;

Constraint function applied to the kernel weights matrix.

property kernelInitializer

kernelInitializer?: InitializerIdentifier | Initializer;

Initializer for the kernel weights matrix, used for the linear transformation of the inputs.

property kernelRegularizer

kernelRegularizer?: RegularizerIdentifier | Regularizer;

Regularizer function applied to the kernel weights matrix.

property recurrentConstraint

recurrentConstraint?: ConstraintIdentifier | Constraint;

Constraint function applied to the recurrentKernel weights matrix.

property recurrentDropout

recurrentDropout?: number;

Number between 0 and 1. Fraction of the units to drop for the linear transformation of the recurrent state.

property recurrentInitializer

recurrentInitializer?: InitializerIdentifier | Initializer;

Initializer for the recurrentKernel weights matrix, used for linear transformation of the recurrent state.

property recurrentRegularizer

recurrentRegularizer?: RegularizerIdentifier | Regularizer;

Regularizer function applied to the recurrentKernel weights matrix.

property units

units: number;

Positive integer, dimensionality of the output space.

property useBias

useBias?: boolean;

Whether the layer uses a bias vector.

Type Aliases

type ClassWeight

type ClassWeight = {
    [classIndex: number]: number;
};

For multi-class classification problems, this object is designed to store a mapping from class index to the "weight" of the class, where higher weighted classes have larger impact on loss, accuracy, and other metrics.
This is useful for cases in which you want the model to "pay more attention" to examples from an under-represented class, e.g., in unbalanced datasets.

type ClassWeightMap

type ClassWeightMap = {
    [outputName: string]: ClassWeight;
};

Class weighting for a model with multiple outputs.
This object maps each output name to a class-weighting object.

type Logs

type Logs = {
    [key: string]: number;
};

Logs in which values can only be numbers.
Used when calling client-provided custom callbacks.

type Shape

type Shape = Array<null | number>;

(null | number)[]

Namespaces

namespace constraints

module 'dist/exports_constraints.d.ts' {}

Copyright 2018 Google LLC
Use of this source code is governed by an MIT-style license that can be found in the LICENSE file or at https://opensource.org/licenses/MIT. =============================================================================

function maxNorm

maxNorm: (args: MaxNormArgs) => Constraint;

MaxNorm weight constraint.
Constrains the weights incident to each hidden unit to have a norm less than or equal to a desired value.
References - [Dropout: A Simple Way to Prevent Neural Networks from Overfitting Srivastava, Hinton, et al. 2014](http://www.cs.toronto.edu/~rsalakhu/papers/srivastava14a.pdf)
{heading: 'Constraints',namespace: 'constraints'}

function minMaxNorm

minMaxNorm: (config: MinMaxNormArgs) => Constraint;

{heading: 'Constraints', namespace: 'constraints'}

function nonNeg

nonNeg: () => Constraint;

Constrains the weight to be non-negative.
{heading: 'Constraints', namespace: 'constraints'}

function unitNorm

unitNorm: (args: UnitNormArgs) => Constraint;

Constrains the weights incident to each hidden unit to have unit norm.
{heading: 'Constraints', namespace: 'constraints'}

namespace initializers

module 'dist/exports_initializers.d.ts' {}

Copyright 2018 Google LLC
Use of this source code is governed by an MIT-style license that can be found in the LICENSE file or at https://opensource.org/licenses/MIT. =============================================================================

function constant

constant: (args: ConstantArgs) => Initializer;

Initializer that generates values initialized to some constant.
{heading: 'Initializers', namespace: 'initializers'}

function glorotNormal

glorotNormal: (args: SeedOnlyInitializerArgs) => Initializer;

Glorot normal initializer, also called Xavier normal initializer. It draws samples from a truncated normal distribution centered on 0 with stddev = sqrt(2 / (fan_in + fan_out)) where fan_in is the number of input units in the weight tensor and fan_out is the number of output units in the weight tensor.
Reference: Glorot & Bengio, AISTATS 2010 http://jmlr.org/proceedings/papers/v9/glorot10a/glorot10a.pdf
{heading: 'Initializers', namespace: 'initializers'}

function glorotUniform

glorotUniform: (args: SeedOnlyInitializerArgs) => Initializer;

Glorot uniform initializer, also called Xavier uniform initializer. It draws samples from a uniform distribution within [-limit, limit] where limit is sqrt(6 / (fan_in + fan_out)) where fan_in is the number of input units in the weight tensor and fan_out is the number of output units in the weight tensor
Reference: Glorot & Bengio, AISTATS 2010 http://jmlr.org/proceedings/papers/v9/glorot10a/glorot10a.pdf.
{heading: 'Initializers', namespace: 'initializers'}

function heNormal

heNormal: (args: SeedOnlyInitializerArgs) => Initializer;

He normal initializer.
It draws samples from a truncated normal distribution centered on 0 with stddev = sqrt(2 / fanIn) where fanIn is the number of input units in the weight tensor.
Reference: He et al., http://arxiv.org/abs/1502.01852
{heading: 'Initializers', namespace: 'initializers'}

function heUniform

heUniform: (args: SeedOnlyInitializerArgs) => Initializer;

He uniform initializer.
It draws samples from a uniform distribution within [-limit, limit] where limit is sqrt(6 / fan_in) where fanIn is the number of input units in the weight tensor.
Reference: He et al., http://arxiv.org/abs/1502.01852
{heading: 'Initializers',namespace: 'initializers'}

function identity

identity: (args: IdentityArgs) => Initializer;

Initializer that generates the identity matrix. Only use for square 2D matrices.
{heading: 'Initializers', namespace: 'initializers'}

function leCunNormal

leCunNormal: (args: SeedOnlyInitializerArgs) => Initializer;

LeCun normal initializer.
It draws samples from a truncated normal distribution centered on 0 with stddev = sqrt(1 / fanIn) where fanIn is the number of input units in the weight tensor.
References: [Self-Normalizing Neural Networks](https://arxiv.org/abs/1706.02515) [Efficient Backprop](http://yann.lecun.com/exdb/publis/pdf/lecun-98b.pdf)
{heading: 'Initializers', namespace: 'initializers'}

function leCunUniform

leCunUniform: (args: SeedOnlyInitializerArgs) => Initializer;

LeCun uniform initializer.
It draws samples from a uniform distribution in the interval [-limit, limit] with limit = sqrt(3 / fanIn), where fanIn is the number of input units in the weight tensor.
{heading: 'Initializers', namespace: 'initializers'}

function ones

ones: () => Initializer;

Initializer that generates tensors initialized to 1.
{heading: 'Initializers', namespace: 'initializers'}

function orthogonal

orthogonal: (args: OrthogonalArgs) => Initializer;

Initializer that generates a random orthogonal matrix.
Reference: [Saxe et al., http://arxiv.org/abs/1312.6120](http://arxiv.org/abs/1312.6120)
{heading: 'Initializers', namespace: 'initializers'}

function randomNormal

randomNormal: (args: RandomNormalArgs) => Initializer;

Initializer that generates random values initialized to a normal distribution.
{heading: 'Initializers', namespace: 'initializers'}

function randomUniform

randomUniform: (args: RandomUniformArgs) => Initializer;

Initializer that generates random values initialized to a uniform distribution.
Values will be distributed uniformly between the configured minval and maxval.
{heading: 'Initializers', namespace: 'initializers'}

function truncatedNormal

truncatedNormal: (args: TruncatedNormalArgs) => Initializer;

Initializer that generates random values initialized to a truncated normal distribution.
These values are similar to values from a RandomNormal except that values more than two standard deviations from the mean are discarded and re-drawn. This is the recommended initializer for neural network weights and filters.
{heading: 'Initializers', namespace: 'initializers'}

function varianceScaling

varianceScaling: (config: VarianceScalingArgs) => Initializer;

Initializer capable of adapting its scale to the shape of weights. With distribution=NORMAL, samples are drawn from a truncated normal distribution centered on zero, with stddev = sqrt(scale / n) where n is: - number of input units in the weight tensor, if mode = FAN_IN. - number of output units, if mode = FAN_OUT. - average of the numbers of input and output units, if mode = FAN_AVG. With distribution=UNIFORM, samples are drawn from a uniform distribution within [-limit, limit], with limit = sqrt(3 * scale / n).
{heading: 'Initializers',namespace: 'initializers'}

function zeros

zeros: () => Zeros;

Initializer that generates tensors initialized to 0.
{heading: 'Initializers', namespace: 'initializers'}

namespace layers

module 'dist/exports_layers.d.ts' {}

Copyright 2018 Google LLC
Use of this source code is governed by an MIT-style license that can be found in the LICENSE file or at https://opensource.org/licenses/MIT. =============================================================================

variable globalMaxPool1d

const globalMaxPool1d: (args?: LayerArgs) => GlobalMaxPooling1D;

variable globalMaxPool2d

const globalMaxPool2d: (args: GlobalPooling2DLayerArgs) => GlobalMaxPooling2D;

variable maxPool1d

const maxPool1d: (args: Pooling1DLayerArgs) => MaxPooling1D;

variable maxPool2d

const maxPool2d: (args: Pooling2DLayerArgs) => MaxPooling2D;

function activation

activation: (args: ActivationLayerArgs) => Activation;

Applies an activation function to an output.

This layer applies element-wise activation function. Other layers, notably dense can also apply activation functions. Use this isolated activation function to extract the values before and after the activation. For instance:

const input = tf.input({shape: [5]});
const denseLayer = tf.layers.dense({units: 1});
const activationLayer = tf.layers.activation({activation: 'relu6'});

// Obtain the output symbolic tensors by applying the layers in order.
const denseOutput = denseLayer.apply(input);
const activationOutput = activationLayer.apply(denseOutput);

// Create the model based on the inputs.
const model = tf.model({
    inputs: input,
    outputs: [denseOutput, activationOutput]
});

// Collect both outputs and print separately.
const [denseOut, activationOut] = model.predict(tf.randomNormal([6, 5]));
denseOut.print();
activationOut.print();

{heading: 'Layers', subheading: 'Basic', namespace: 'layers'}

function add

add: (args?: LayerArgs) => Add;

Layer that performs element-wise addition on an Array of inputs.
It takes as input a list of tensors, all of the same shape, and returns a single tensor (also of the same shape). The inputs are specified as an Array when the apply method of the Add layer instance is called. For example:
const input1 = tf.input({shape: [2, 2]}); const input2 = tf.input({shape: [2, 2]}); const addLayer = tf.layers.add(); const sum = addLayer.apply([input1, input2]); console.log(JSON.stringify(sum.shape)); // You get [null, 2, 2], with the first dimension as the undetermined batch // dimension.
{heading: 'Layers', subheading: 'Merge', namespace: 'layers'}

function alphaDropout

alphaDropout: (args: AlphaDropoutArgs) => AlphaDropout;

Applies Alpha Dropout to the input.
As it is a regularization layer, it is only active at training time.
Alpha Dropout is a Dropout that keeps mean and variance of inputs to their original values, in order to ensure the self-normalizing property even after this dropout. Alpha Dropout fits well to Scaled Exponential Linear Units by randomly setting activations to the negative saturation value.
Arguments: - rate: float, drop probability (as with Dropout). The multiplicative noise will have standard deviation sqrt(rate / (1 - rate)). - noise_shape: A 1-D Tensor of type int32, representing the shape for randomly generated keep/drop flags.
Input shape: Arbitrary. Use the keyword argument inputShape (tuple of integers, does not include the samples axis) when using this layer as the first layer in a model.
Output shape: Same shape as input.
References: - [Self-Normalizing Neural Networks](https://arxiv.org/abs/1706.02515)
{heading: 'Layers', subheading: 'Noise', namespace: 'layers'}

function average

average: (args?: LayerArgs) => Average;

Layer that performs element-wise averaging on an Array of inputs.

It takes as input a list of tensors, all of the same shape, and returns a single tensor (also of the same shape). For example:

const input1 = tf.input({shape: [2, 2]});
const input2 = tf.input({shape: [2, 2]});
const averageLayer = tf.layers.average();
const average = averageLayer.apply([input1, input2]);
console.log(JSON.stringify(average.shape));
// You get [null, 2, 2], with the first dimension as the undetermined batch
// dimension.

{heading: 'Layers', subheading: 'Merge', namespace: 'layers'}

function averagePooling1d

averagePooling1d: (args: Pooling1DLayerArgs) => AveragePooling1D;

Average pooling operation for spatial data.
Input shape: [batchSize, inLength, channels]
Output shape: [batchSize, pooledLength, channels]
tf.avgPool1d is an alias.
{heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}

function averagePooling2d

averagePooling2d: (args: Pooling2DLayerArgs) => AveragePooling2D;

Average pooling operation for spatial data.
Input shape: - If dataFormat === CHANNEL_LAST: 4D tensor with shape: [batchSize, rows, cols, channels] - If dataFormat === CHANNEL_FIRST: 4D tensor with shape: [batchSize, channels, rows, cols]
Output shape - If dataFormat === CHANNEL_LAST: 4D tensor with shape: [batchSize, pooledRows, pooledCols, channels] - If dataFormat === CHANNEL_FIRST: 4D tensor with shape: [batchSize, channels, pooledRows, pooledCols]
tf.avgPool2d is an alias.
{heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}

function averagePooling3d

averagePooling3d: (args: Pooling3DLayerArgs) => AveragePooling3D;

Average pooling operation for 3D data.
Input shape - If dataFormat === channelsLast: 5D tensor with shape: [batchSize, depths, rows, cols, channels] - If dataFormat === channelsFirst: 4D tensor with shape: [batchSize, channels, depths, rows, cols]
Output shape - If dataFormat=channelsLast: 5D tensor with shape: [batchSize, pooledDepths, pooledRows, pooledCols, channels] - If dataFormat=channelsFirst: 5D tensor with shape: [batchSize, channels, pooledDepths, pooledRows, pooledCols]
{heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}

function avgPool1d

avgPool1d: (args: Pooling1DLayerArgs) => AveragePooling1D;

function avgPool2d

avgPool2d: (args: Pooling2DLayerArgs) => AveragePooling2D;

function avgPool3d

avgPool3d: (args: Pooling3DLayerArgs) => AveragePooling3D;

function avgPooling1d

avgPooling1d: (args: Pooling1DLayerArgs) => AveragePooling1D;

function avgPooling2d

avgPooling2d: (args: Pooling2DLayerArgs) => AveragePooling2D;

function avgPooling3d

avgPooling3d: (args: Pooling3DLayerArgs) => AveragePooling3D;

function batchNormalization

batchNormalization: (args?: BatchNormalizationLayerArgs) => BatchNormalization;

Batch normalization layer (Ioffe and Szegedy, 2014).
Normalize the activations of the previous layer at each batch, i.e. applies a transformation that maintains the mean activation close to 0 and the activation standard deviation close to 1.
Input shape: Arbitrary. Use the keyword argument inputShape (Array of integers, does not include the sample axis) when calling the constructor of this class, if this layer is used as a first layer in a model.
Output shape: Same shape as input.
References: - [Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift](https://arxiv.org/abs/1502.03167)
{heading: 'Layers', subheading: 'Normalization', namespace: 'layers'}

function bidirectional

bidirectional: (args: BidirectionalLayerArgs) => Bidirectional;

{heading: 'Layers', subheading: 'Wrapper', namespace: 'layers'}

function categoryEncoding

categoryEncoding: (args: CategoryEncodingArgs) => CategoryEncoding;

A preprocessing layer which encodes integer features.
This layer provides options for condensing data into a categorical encoding when the total number of tokens are known in advance. It accepts integer values as inputs, and it outputs a dense representation of those inputs.
Arguments:
numTokens: The total number of tokens the layer should support. All inputs to the layer must integers in the range `0 <= value < numTokens`, or an error will be thrown.
outputMode: Specification for the output of the layer. Defaults to multiHot. Values can be oneHot, multiHot or count, configuring the layer as follows:
oneHot: Encodes each individual element in the input into an array of numTokens size, containing a 1 at the element index. If the last dimension is size 1, will encode on that dimension. If the last dimension is not size 1, will append a new dimension for the encoded output.
multiHot: Encodes each sample in the input into a single array of numTokens size, containing a 1 for each vocabulary term present in the sample. Treats the last dimension as the sample dimension, if input shape is (..., sampleLength), output shape will be (..., numTokens).
count: Like multiHot, but the int array contains a count of the number of times the token at that index appeared in the sample.
For all output modes, currently only output up to rank 2 is supported. Call arguments: inputs: A 1D or 2D tensor of integer inputs. countWeights: A tensor in the same shape as inputs indicating the weight for each sample value when summing up in count mode. Not used in multiHot or oneHot modes.
{heading: 'Layers', subheading: 'CategoryEncoding', namespace: 'layers'}

function centerCrop

centerCrop: (args?: CenterCropArgs) => CenterCrop;

A preprocessing layer which center crops images.
This layers crops the central portion of the images to a target size. If an image is smaller than the target size, it will be resized and cropped so as to return the largest possible window in the image that matches the target aspect ratio.
Input pixel values can be of any range (e.g. [0., 1.) or [0, 255]) and of integer or floating point dtype.
If the input height/width is even and the target height/width is odd (or inversely), the input image is left-padded by 1 pixel.
Arguments: height: Integer, the height of the output shape. width: Integer, the width of the output shape.
Input shape: 3D (unbatched) or 4D (batched) tensor with shape: (..., height, width, channels), in channelsLast format.
Output shape: 3D (unbatched) or 4D (batched) tensor with shape: (..., targetHeight, targetWidth, channels).
{heading: 'Layers', subheading: 'CenterCrop', namespace: 'layers'}

function concatenate

concatenate: (args?: ConcatenateLayerArgs) => Concatenate;

Layer that concatenates an Array of inputs.

It takes a list of tensors, all of the same shape except for the concatenation axis, and returns a single tensor, the concatenation of all inputs. For example:

const input1 = tf.input({shape: [2, 2]});
const input2 = tf.input({shape: [2, 3]});
const concatLayer = tf.layers.concatenate();
const output = concatLayer.apply([input1, input2]);
console.log(JSON.stringify(output.shape));
// You get [null, 2, 5], with the first dimension as the undetermined batch
// dimension. The last dimension (5) is the result of concatenating the
// last dimensions of the inputs (2 and 3).

{heading: 'Layers', subheading: 'Merge', namespace: 'layers'}

function conv1d

conv1d: (args: ConvLayerArgs) => Conv1D;

1D convolution layer (e.g., temporal convolution).
This layer creates a convolution kernel that is convolved with the layer input over a single spatial (or temporal) dimension to produce a tensor of outputs.
If use_bias is True, a bias vector is created and added to the outputs.
If activation is not null, it is applied to the outputs as well.
When using this layer as the first layer in a model, provide an inputShape argument Array or null.
For example, inputShape would be: - [10, 128] for sequences of 10 vectors of 128-dimensional vectors - [null, 128] for variable-length sequences of 128-dimensional vectors.
{heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}

function conv2d

conv2d: (args: ConvLayerArgs) => Conv2D;

2D convolution layer (e.g. spatial convolution over images).
This layer creates a convolution kernel that is convolved with the layer input to produce a tensor of outputs.
If useBias is True, a bias vector is created and added to the outputs.
If activation is not null, it is applied to the outputs as well.
When using this layer as the first layer in a model, provide the keyword argument inputShape (Array of integers, does not include the sample axis), e.g. inputShape=[128, 128, 3] for 128x128 RGB pictures in dataFormat='channelsLast'.
{heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}

function conv2dTranspose

conv2dTranspose: (args: ConvLayerArgs) => Conv2DTranspose;

Transposed convolutional layer (sometimes called Deconvolution).
The need for transposed convolutions generally arises from the desire to use a transformation going in the opposite direction of a normal convolution, i.e., from something that has the shape of the output of some convolution to something that has the shape of its input while maintaining a connectivity pattern that is compatible with said convolution.
When using this layer as the first layer in a model, provide the configuration inputShape (Array of integers, does not include the sample axis), e.g., inputShape: [128, 128, 3] for 128x128 RGB pictures in dataFormat: 'channelsLast'.
Input shape: 4D tensor with shape: [batch, channels, rows, cols] if dataFormat is 'channelsFirst'. or 4D tensor with shape [batch, rows, cols, channels] if dataFormat is 'channelsLast'.
Output shape: 4D tensor with shape: [batch, filters, newRows, newCols] if dataFormat is 'channelsFirst'. or 4D tensor with shape: [batch, newRows, newCols, filters] if dataFormat is 'channelsLast'.
References: - [A guide to convolution arithmetic for deep learning](https://arxiv.org/abs/1603.07285v1) - [Deconvolutional Networks](http://www.matthewzeiler.com/pubs/cvpr2010/cvpr2010.pdf)
{heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}

function conv3d

conv3d: (args: ConvLayerArgs) => Conv3D;

3D convolution layer (e.g. spatial convolution over volumes).
This layer creates a convolution kernel that is convolved with the layer input to produce a tensor of outputs.
If useBias is True, a bias vector is created and added to the outputs.
If activation is not null, it is applied to the outputs as well.
When using this layer as the first layer in a model, provide the keyword argument inputShape (Array of integers, does not include the sample axis), e.g. inputShape=[128, 128, 128, 1] for 128x128x128 grayscale volumes in dataFormat='channelsLast'.
{heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}

function conv3dTranspose

conv3dTranspose: (args: ConvLayerArgs) => Layer;

function convLstm2d

convLstm2d: (args: ConvLSTM2DArgs) => ConvLSTM2D;

{heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}

function convLstm2dCell

convLstm2dCell: (args: ConvLSTM2DCellArgs) => ConvLSTM2DCell;

{heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}

function cropping2D

cropping2D: (args: Cropping2DLayerArgs) => Cropping2D;

Cropping layer for 2D input (e.g., image).
This layer can crop an input at the top, bottom, left and right side of an image tensor.
Input shape: 4D tensor with shape: - If dataFormat is "channelsLast": [batch, rows, cols, channels] - If data_format is "channels_first": [batch, channels, rows, cols].
Output shape: 4D with shape: - If dataFormat is "channelsLast": [batch, croppedRows, croppedCols, channels] - If dataFormat is "channelsFirst": [batch, channels, croppedRows, croppedCols].
Examples
const model = tf.sequential(); model.add(tf.layers.cropping2D({cropping:[[2, 2], [2, 2]], inputShape: [128, 128, 3]})); //now output shape is [batch, 124, 124, 3]
{heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}

function dense

dense: (args: DenseLayerArgs) => Dense;

Creates a dense (fully connected) layer.
This layer implements the operation: output = activation(dot(input, kernel) + bias)
activation is the element-wise activation function passed as the activation argument.
kernel is a weights matrix created by the layer.
bias is a bias vector created by the layer (only applicable if useBias is true).
**Input shape:**
nD tf.Tensor with shape: (batchSize, ..., inputDim).
The most common situation would be a 2D input with shape (batchSize, inputDim).
**Output shape:**
nD tensor with shape: (batchSize, ..., units).
For instance, for a 2D input with shape (batchSize, inputDim), the output would have shape (batchSize, units).
Note: if the input to the layer has a rank greater than 2, then it is flattened prior to the initial dot product with the kernel.
{heading: 'Layers', subheading: 'Basic', namespace: 'layers'}

function depthwiseConv2d

depthwiseConv2d: (args: DepthwiseConv2DLayerArgs) => DepthwiseConv2D;

Depthwise separable 2D convolution.
Depthwise Separable convolutions consists in performing just the first step in a depthwise spatial convolution (which acts on each input channel separately). The depthMultiplier argument controls how many output channels are generated per input channel in the depthwise step.
{heading: 'Layers', subheading: 'Convolutional', namespace: 'layers'}

function dot

dot: (args: DotLayerArgs) => Dot;

Layer that computes a dot product between samples in two tensors.
E.g., if applied to a list of two tensors a and b both of shape [batchSize, n], the output will be a tensor of shape [batchSize, 1], where each entry at index [i, 0] will be the dot product between a[i, :] and b[i, :].
Example:
const dotLayer = tf.layers.dot({axes: -1}); const x1 = tf.tensor2d([[10, 20], [30, 40]]); const x2 = tf.tensor2d([[-1, -2], [-3, -4]]); // Invoke the layer's apply() method in eager (imperative) mode. const y = dotLayer.apply([x1, x2]); y.print();
{heading: 'Layers', subheading: 'Merge', namespace: 'layers'}

function dropout

dropout: (args: DropoutLayerArgs) => Dropout;

Applies [dropout](http://www.cs.toronto.edu/~rsalakhu/papers/srivastava14a.pdf) to the input.
Dropout consists in randomly setting a fraction rate of input units to 0 at each update during training time, which helps prevent overfitting.
{heading: 'Layers', subheading: 'Basic', namespace: 'layers'}

function elu

elu: (args?: ELULayerArgs) => ELU;

Exponential Linear Unit (ELU).
It follows: f(x) = alpha * (exp(x) - 1.) for x < 0, f(x) = x for x >= 0.
Input shape: Arbitrary. Use the configuration inputShape when using this layer as the first layer in a model.
Output shape: Same shape as the input.
References: - [Fast and Accurate Deep Network Learning by Exponential Linear Units (ELUs)](https://arxiv.org/abs/1511.07289v1)
{ heading: 'Layers', subheading: 'Advanced Activation', namespace: 'layers' }

function embedding

embedding: (args: EmbeddingLayerArgs) => Embedding;

Maps positive integers (indices) into dense vectors of fixed size. E.g. [[4], [20]] -> [[0.25, 0.1], [0.6, -0.2]]
**Input shape:** 2D tensor with shape: [batchSize, sequenceLength].
**Output shape:** 3D tensor with shape: `[batchSize, sequenceLength, outputDim]`.
{heading: 'Layers', subheading: 'Basic', namespace: 'layers'}

function flatten

flatten: (args?: FlattenLayerArgs) => Flatten;

Flattens the input. Does not affect the batch size.

A Flatten layer flattens each batch in its inputs to 1D (making the output 2D).

For example:

const input = tf.input({shape: [4, 3]});
const flattenLayer = tf.layers.flatten();
// Inspect the inferred output shape of the flatten layer, which
// equals `[null, 12]`. The 2nd dimension is 4 * 3, i.e., the result of the
// flattening. (The 1st dimension is the undermined batch size.)
console.log(JSON.stringify(flattenLayer.apply(input).shape));

{heading: 'Layers', subheading: 'Basic', namespace: 'layers'}

function gaussianDropout

gaussianDropout: (args: GaussianDropoutArgs) => GaussianDropout;

Apply multiplicative 1-centered Gaussian noise.
As it is a regularization layer, it is only active at training time.
Arguments: - rate: float, drop probability (as with Dropout). The multiplicative noise will have standard deviation sqrt(rate / (1 - rate)).
Input shape: Arbitrary. Use the keyword argument inputShape (tuple of integers, does not include the samples axis) when using this layer as the first layer in a model.
Output shape: Same shape as input.
References: - [Dropout: A Simple Way to Prevent Neural Networks from Overfitting]( http://www.cs.toronto.edu/~rsalakhu/papers/srivastava14a.pdf)
{heading: 'Layers', subheading: 'Noise', namespace: 'layers'}

function gaussianNoise

gaussianNoise: (args: GaussianNoiseArgs) => GaussianNoise;

Apply additive zero-centered Gaussian noise.
As it is a regularization layer, it is only active at training time.
This is useful to mitigate overfitting (you could see it as a form of random data augmentation). Gaussian Noise (GS) is a natural choice as corruption process for real valued inputs.
# Arguments stddev: float, standard deviation of the noise distribution.
# Input shape Arbitrary. Use the keyword argument input_shape (tuple of integers, does not include the samples axis) when using this layer as the first layer in a model.
# Output shape Same shape as input.
{heading: 'Layers', subheading: 'Noise', namespace: 'layers'}

function globalAveragePooling1d

globalAveragePooling1d: (args?: LayerArgs) => GlobalAveragePooling1D;

Global average pooling operation for temporal data.
Input Shape: 3D tensor with shape: [batchSize, steps, features].
Output Shape: 2D tensor with shape: [batchSize, features].
{heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}

function globalAveragePooling2d

globalAveragePooling2d: (
    args: GlobalPooling2DLayerArgs
) => GlobalAveragePooling2D;

Global average pooling operation for spatial data.
Input shape: - If dataFormat is CHANNEL_LAST: 4D tensor with shape: [batchSize, rows, cols, channels]. - If dataFormat is CHANNEL_FIRST: 4D tensor with shape: [batchSize, channels, rows, cols].
Output shape: 2D tensor with shape: [batchSize, channels].
{heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}

function globalMaxPooling1d

globalMaxPooling1d: (args?: LayerArgs) => GlobalMaxPooling1D;

Global max pooling operation for temporal data.
Input Shape: 3D tensor with shape: [batchSize, steps, features].
Output Shape: 2D tensor with shape: [batchSize, features].
{heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}

function globalMaxPooling2d

globalMaxPooling2d: (args: GlobalPooling2DLayerArgs) => GlobalMaxPooling2D;

Global max pooling operation for spatial data.
Input shape: - If dataFormat is CHANNEL_LAST: 4D tensor with shape: [batchSize, rows, cols, channels]. - If dataFormat is CHANNEL_FIRST: 4D tensor with shape: [batchSize, channels, rows, cols].
Output shape: 2D tensor with shape: [batchSize, channels].
{heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}

function gru

gru: (args: GRULayerArgs) => GRU;

Gated Recurrent Unit - Cho et al. 2014.
This is an RNN layer consisting of one GRUCell. However, unlike the underlying GRUCell, the apply method of SimpleRNN operates on a sequence of inputs. The shape of the input (not including the first, batch dimension) needs to be at least 2-D, with the first dimension being time steps. For example:
```js const rnn = tf.layers.gru({units: 8, returnSequences: true});
// Create an input with 10 time steps. const input = tf.input({shape: [10, 20]}); const output = rnn.apply(input);
console.log(JSON.stringify(output.shape)); // [null, 10, 8]: 1st dimension is unknown batch size; 2nd dimension is the // same as the sequence length of input, due to returnSequences: true; // 3rd dimension is the GRUCell's number of units.
{heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}

function gruCell

gruCell: (args: GRUCellLayerArgs) => GRUCell;

Cell class for GRU.

GRUCell is distinct from the RNN subclass GRU in that its apply method takes the input data of only a single time step and returns the cell's output at the time step, while GRU takes the input data over a number of time steps. For example:

const cell = tf.layers.gruCell({units: 2});
const input = tf.input({shape: [10]});
const output = cell.apply(input);

console.log(JSON.stringify(output.shape));
// [null, 10]: This is the cell's output at a single time step. The 1st
// dimension is the unknown batch size.

Instance(s) of GRUCell can be used to construct RNN layers. The most typical use of this workflow is to combine a number of cells into a stacked RNN cell (i.e., StackedRNNCell internally) and use it to create an RNN. For example:

const cells = [
  tf.layers.gruCell({units: 4}),
  tf.layers.gruCell({units: 8}),
];
const rnn = tf.layers.rnn({cell: cells, returnSequences: true});

// Create an input with 10 time steps and a length-20 vector at each step.
const input = tf.input({shape: [10, 20]});
const output = rnn.apply(input);

console.log(JSON.stringify(output.shape));
// [null, 10, 8]: 1st dimension is unknown batch size; 2nd dimension is the
// same as the sequence length of `input`, due to `returnSequences`: `true`;
// 3rd dimension is the last `gruCell`'s number of units.

To create an RNN consisting of only *one* GRUCell, use the tf.layers.gru.

{heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}

function input

input: (config: InputConfig) => SymbolicTensor;

Used to instantiate an input to a model as a tf.SymbolicTensor.
Users should call the input factory function for consistency with other generator functions.
Example:
// Defines a simple logistic regression model with 32 dimensional input // and 3 dimensional output. const x = tf.input({shape: [32]}); const y = tf.layers.dense({units: 3, activation: 'softmax'}).apply(x); const model = tf.model({inputs: x, outputs: y}); model.predict(tf.ones([2, 32])).print();
Note: input is only necessary when using model. When using sequential, specify inputShape for the first layer or use inputLayer as the first layer.
{heading: 'Models', subheading: 'Inputs'}

function inputLayer

inputLayer: (args: InputLayerArgs) => InputLayer;

An input layer is an entry point into a tf.LayersModel.

InputLayer is generated automatically for tf.Sequential models by specifying the inputshape or batchInputShape for the first layer. It should not be specified explicitly. However, it can be useful sometimes, e.g., when constructing a sequential model from a subset of another sequential model's layers. Like the code snippet below shows.

// Define a model which simply adds two inputs.
const model1 = tf.sequential();
model1.add(tf.layers.dense({inputShape: [4], units: 3, activation: 'relu'}));
model1.add(tf.layers.dense({units: 1, activation: 'sigmoid'}));
model1.summary();
model1.predict(tf.zeros([1, 4])).print();

// Construct another model, reusing the second layer of `model1` while
// not using the first layer of `model1`. Note that you cannot add the second
// layer of `model` directly as the first layer of the new sequential model,
// because doing so will lead to an error related to the fact that the layer
// is not an input layer. Instead, you need to create an `inputLayer` and add
// it to the new sequential model before adding the reused layer.
const model2 = tf.sequential();
// Use an inputShape that matches the input shape of `model1`'s second
// layer.
model2.add(tf.layers.inputLayer({inputShape: [3]}));
model2.add(model1.layers[1]);
model2.summary();
model2.predict(tf.zeros([1, 3])).print();

{heading: 'Layers', subheading: 'Inputs', namespace: 'layers'}

function layerNormalization

layerNormalization: (args?: LayerNormalizationLayerArgs) => LayerNormalization;

Layer-normalization layer (Ba et al., 2016).
Normalizes the activations of the previous layer for each given example in a batch independently, instead of across a batch like in batchNormalization. In other words, this layer applies a transformation that maintains the mean activation within each example close to 0 and activation variance close to 1.
Input shape: Arbitrary. Use the argument inputShape when using this layer as the first layer in a model.
Output shape: Same as input.
References: - [Layer Normalization](https://arxiv.org/abs/1607.06450)
{heading: 'Layers', subheading: 'Normalization', namespace: 'layers'}

function leakyReLU

leakyReLU: (args?: LeakyReLULayerArgs) => LeakyReLU;

Leaky version of a rectified linear unit.
It allows a small gradient when the unit is not active: f(x) = alpha * x for x < 0. f(x) = x for x >= 0.
Input shape: Arbitrary. Use the configuration inputShape when using this layer as the first layer in a model.
Output shape: Same shape as the input.
{ heading: 'Layers', subheading: 'Advanced Activation', namespace: 'layers' }

function lstm

lstm: (args: LSTMLayerArgs) => LSTM;

Long-Short Term Memory layer - Hochreiter 1997.
This is an RNN layer consisting of one LSTMCell. However, unlike the underlying LSTMCell, the apply method of LSTM operates on a sequence of inputs. The shape of the input (not including the first, batch dimension) needs to be at least 2-D, with the first dimension being time steps. For example:
```js const lstm = tf.layers.lstm({units: 8, returnSequences: true});
// Create an input with 10 time steps. const input = tf.input({shape: [10, 20]}); const output = lstm.apply(input);
console.log(JSON.stringify(output.shape)); // [null, 10, 8]: 1st dimension is unknown batch size; 2nd dimension is the // same as the sequence length of input, due to returnSequences: true; // 3rd dimension is the LSTMCell's number of units.
{heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}

function lstmCell

lstmCell: (args: LSTMCellLayerArgs) => LSTMCell;

Cell class for LSTM.

LSTMCell is distinct from the RNN subclass LSTM in that its apply method takes the input data of only a single time step and returns the cell's output at the time step, while LSTM takes the input data over a number of time steps. For example:

const cell = tf.layers.lstmCell({units: 2});
const input = tf.input({shape: [10]});
const output = cell.apply(input);

console.log(JSON.stringify(output.shape));
// [null, 10]: This is the cell's output at a single time step. The 1st
// dimension is the unknown batch size.

Instance(s) of LSTMCell can be used to construct RNN layers. The most typical use of this workflow is to combine a number of cells into a stacked RNN cell (i.e., StackedRNNCell internally) and use it to create an RNN. For example:

const cells = [
  tf.layers.lstmCell({units: 4}),
  tf.layers.lstmCell({units: 8}),
];
const rnn = tf.layers.rnn({cell: cells, returnSequences: true});

// Create an input with 10 time steps and a length-20 vector at each step.
const input = tf.input({shape: [10, 20]});
const output = rnn.apply(input);

console.log(JSON.stringify(output.shape));
// [null, 10, 8]: 1st dimension is unknown batch size; 2nd dimension is the
// same as the sequence length of `input`, due to `returnSequences`: `true`;
// 3rd dimension is the last `lstmCell`'s number of units.

To create an RNN consisting of only *one* LSTMCell, use the tf.layers.lstm.

{heading: 'Layers', subheading: 'Recurrent', namespace: 'layers'}

function masking

masking: (args?: MaskingArgs) => Masking;

Masks a sequence by using a mask value to skip timesteps.
If all features for a given sample timestep are equal to mask_value, then the sample timestep will be masked (skipped) in all downstream layers (as long as they support masking).
If any downstream layer does not support masking yet receives such an input mask, an exception will be raised.
Arguments: - maskValue: Either None or mask value to skip.
Input shape: Arbitrary. Use the keyword argument inputShape (tuple of integers, does not include the samples axis) when using this layer as the first layer in a model.
Output shape: Same shape as input.
{heading: 'Layers', subheading: 'Mask', namespace: 'layers'}

function maximum

maximum: (args?: LayerArgs) => Maximum;

Layer that computes the element-wise maximum of an Array of inputs.

It takes as input a list of tensors, all of the same shape, and returns a single tensor (also of the same shape). For example:

const input1 = tf.input({shape: [2, 2]});
const input2 = tf.input({shape: [2, 2]});
const maxLayer = tf.layers.maximum();
const max = maxLayer.apply([input1, input2]);
console.log(JSON.stringify(max.shape));
// You get [null, 2, 2], with the first dimension as the undetermined batch
// dimension.

{heading: 'Layers', subheading: 'Merge', namespace: 'layers'}

function maxPooling1d

maxPooling1d: (args: Pooling1DLayerArgs) => MaxPooling1D;

Max pooling operation for temporal data.
Input shape: [batchSize, inLength, channels]
Output shape: [batchSize, pooledLength, channels]
{heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}

function maxPooling2d

maxPooling2d: (args: Pooling2DLayerArgs) => MaxPooling2D;

Max pooling operation for spatial data.
Input shape - If dataFormat === CHANNEL_LAST: 4D tensor with shape: [batchSize, rows, cols, channels] - If dataFormat === CHANNEL_FIRST: 4D tensor with shape: [batchSize, channels, rows, cols]
Output shape - If dataFormat=CHANNEL_LAST: 4D tensor with shape: [batchSize, pooledRows, pooledCols, channels] - If dataFormat=CHANNEL_FIRST: 4D tensor with shape: [batchSize, channels, pooledRows, pooledCols]
{heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}

function maxPooling3d

maxPooling3d: (args: Pooling3DLayerArgs) => MaxPooling3D;

Max pooling operation for 3D data.
Input shape - If dataFormat === channelsLast: 5D tensor with shape: [batchSize, depths, rows, cols, channels] - If dataFormat === channelsFirst: 5D tensor with shape: [batchSize, channels, depths, rows, cols]
Output shape - If dataFormat=channelsLast: 5D tensor with shape: [batchSize, pooledDepths, pooledRows, pooledCols, channels] - If dataFormat=channelsFirst: 5D tensor with shape: [batchSize, channels, pooledDepths, pooledRows, pooledCols]
{heading: 'Layers', subheading: 'Pooling', namespace: 'layers'}

function minimum

minimum: (args?: LayerArgs) => Minimum;

Layer that computes the element-wise minimum of an Array of inputs.

It takes as input a list of tensors, all of the same shape, and returns a single tensor (also of the same shape). For example:

const input1 = tf.input({shape: [2, 2]});
const input2 = tf.input({shape: [2, 2]});
const minLayer = tf.layers.minimum();
const min = minLayer.apply([input1, input2]);
console.log(JSON.stringify(min.shape));
// You get [null, 2, 2], with the first dimension as the undetermined batch
// dimension.

{heading: 'Layers', subheading: 'Merge', namespace: 'layers'}

function multiply

multiply: (

@tensorflow/tfjs-layers

Install

Overview

Index

Variables

Functions

Classes

Interfaces

Type Aliases

Namespaces

Variables

variable callbacks

variable version_layers

Functions

function input

function loadLayersModel

Parameter pathOrIOHandler

Parameter options

Returns

function model

function registerCallbackConstructor

function sequential

Classes

class Callback

property model

method setModel

class CallbackList

constructor

Parameter callbacks

Parameter queueLength

property callbacks

property queueLength

method append

method onBatchBegin

Parameter batch

Parameter logs

method onBatchEnd

Parameter batch

Parameter logs

method onEpochBegin

Parameter epoch

Parameter logs

method onEpochEnd

Parameter epoch

Parameter logs

method onTrainBegin

Parameter logs

method onTrainEnd

Parameter logs

method setModel

method setParams

class CustomCallback

constructor

property batchBegin

property batchEnd

property epochBegin

property epochEnd

property nextFrameFunc

property nowFunc

property trainBegin

property trainEnd

property yield

method maybeWait

method onBatchBegin

method onBatchEnd

method onEpochBegin

method onEpochEnd

method onTrainBegin

method onTrainEnd

class EarlyStopping

constructor

property baseline

property minDelta

property mode

property monitor

property monitorFunc

property patience

property verbose

method onEpochEnd

method onTrainBegin