Multi-layer Perceptron Regressor

MLPRegressor is a class for fitting a regression task using multi-layer perceptron neural network

Constructor

Import

import * as Datacook from 'datacook';
const { MLPRegressor } = DataCook.Model;

Constructor

const mlp = new MLPRegressor({ hiddenLayerSizes: [ 5, 6, 7 ], activations: 'relu' });

Option parameters

parameter	type	description
hiddenLayerSizes	number[]	size of nodes in hidden layers, length of array represents number of hidden layers in the network
activations	{ ‘relu’ \| ‘sigmoid’ \| ‘tanh’ } or Array of { ‘relu’ \| ‘sigmoid’ \| ‘tanh’ }	activation function of hidden layers, default is ‘sigmoid’
optimizerType	‘sgd’ \| ‘momentum’ \| ‘adagrad’ \| ‘adadelta’ \| ‘adam’ \| ‘adamax’ \| ‘rmsprop’	optimizer types for training. All of the following optimizers types supported in tensorflow.js can be applied. Default to ‘adam’
optimizerProps	OptimizerProps	parameters used to init corresponding optimizer, you can refer to documentations in tensorflow.js to find the supported initailization paratemters for a given type of optimizer. For example, `{ learningRate: 0.1, beta1: 0.1, beta2: 0.2, epsilon: 0.1 }` could be used to initialize adam optimizer.

Methods

fit

Fit model according to X, y.

async fit(xData: Tensor | RecursiveArray<number>,
    yData: Tensor | RecursiveArray<number>,
    params)

Parameters

parameter	type	description
xData	Tensor \| RecursiveArray	Tensor like of shape (n_samples, n_features), input feature
yData	Tensor \| RecursiveArray	Tensor like of shape (n_sample, ), input target values
params	MLPRegressorTrainParams	option in params: - batchSize: batch size: default to 32, - epochs: epochs for training, default to Math.ceil(10000 / ( nData / batchSize ))

predict

Make predictions using logistic regression model.

async predict(xData: Tensor | RecursiveArray<number>): Promise<Tensor>

Parameters

parameter	type	description
xData	Tensor	RecursiveArray	Input features

fromJson

Load model paramters from json string object

async fromJson(modelJson: string)

Parameters

parameter	type	description
modelJson	string	model json string

Examples

import '@tensorflow/tfjs-backend-cpu';
import * as tf from '@tensorflow/tfjs-core';
import * as DataCook from '@pipcook/datacook';
const MLPRegressor = DataCook.Model;
const cases = tf.mul(tf.randomNormal([ 10000, 5 ]), [ 1, 10, 100, 2, 3 ]);
const weight = tf.tensor([ 2, 3, 1, 4, 6 ]);
const y = tf.add(tf.sum(tf.mul(cases, weight), 1), 10);
const mlp = new MLPRegressor({
    hiddenLayerSizes: [ 4, 10 ],
    activations: 'relu',
    optimizerType: 'adam',
    optimizerProps: { learningRate: 0.01 }
});

// model fit
await mlp.fit(cases, y);
// predict
const yPred = await mlp.predict(cases) as Tensor1D;

// save and load
const modelJson = await mlp.toJson();
const mlp2 = new MLPRegressor({});
mlp2.fromJson(modelJson);
const yPred2 = await mlp2.predict(cases);

toJson

Dump model parameters to json string.

async toJson(): Promise<string>