Class: Trainer

Constructor

new Trainer(optionsopt)

Parameters:

Name Type Attributes Description

options

object

Properties

Name	Type	Attributes	Default	Description
`batch`	boolean \| number	<optional>		Use batch, online (stochastic), or mini-batch learning modes. Batch `true`: Connection weights are updated once after iterating through all the training samples in the training data (an epoch). Online `false`: Connection weights are updated after every training sample in the training data. Mini-batch `<number>`: Connection weights are updated every `<number>` training samples.
`errorThreshold`	number	<optional>	0.001	The target `error` value. The goal of the Trainer is to train the Network until the `error` is below this value.
`frequency`	number	<optional>	100	How many iterations through the training data between calling `options.onProgress`.
`maxEpochs`	number	<optional>	20000	The max training iterations. The Trainer will stop training after iterating through the training data this number of times. One full loop through the training data is counted as one epoch.
`onFail`	Trainer~onFail	<optional>		Called if the Network `error` does not fall below the `errorThreshold` after `maxEpochs`.
`onProgress`	Trainer~onProgress	<optional>		Called every `frequency` epochs.
`onSuccess`	Trainer~onSuccess	<optional>		Called if the Network `error` falls below the `errorThreshold` during training.

Source:

Trainer.js, line 16

Example

const network = new anny.Network([2, 1])
const trainer = new Trainer()

trainer.train(network, anny.DATA.ORGate)

network.activate([0, 0]) // => 0.000836743108
network.activate([0, 1]) // => 0.998253857294

Methods

train(network, data, overridesopt)

Train the network to classify the data.

Parameters:

Name Type Attributes Description

network

Network

The Network to be trained.

data

Array.<object>

Array of objects in the form {input: [], output: []}.

overrides

Object

Overrides are merged into this trainer

Properties

Name	Type	Attributes	Description
`onSuccess`	Trainer~onSuccess	<optional>	Overrides are merged into this trainer instance's options.

Source:

Trainer.js, line 74

See:

Network
Data

Type Definitions

onFail(error, epoch)

Called if the network error is not below the errorThreshold after maxEpochs iterations through the training data set.

Parameters:

Name	Type	Description
`error`	number	The `network` error value at the time of success.
`epoch`	number	Indicates on which iteration through the training data the `network` became successful.

Source:

Trainer.js, line 147

onProgress(error, epoch)

Called if the network error falls below the errorThreshold.

Parameters:

Name	Type	Description
`error`	number	The `network` error value at the time of success.
`epoch`	number	Indicates on which iteration through the training data the `network` became successful.

Source:

Trainer.js, line 158

onSuccess(error, epoch)

Called if the network error falls below the errorThreshold.

Parameters:

Name	Type	Description
`error`	number	The `network` error value at the time of success.
`epoch`	number	Indicates on which iteration through the training data the `network` became successful.

Source:

Trainer.js, line 137