test to validate sequence-to-sequence learning.

test_conv3d.js

// Importing necessary libraries: lodash, TensorFlow.js, and colors
let _ = require('lodash');
let tf = require('@tensorflow/tfjs-node');
let colors=require('colors');

// Function to generate a dataset of binary arrays with a given size
function buildDataset(sp) {
    let s = []
    for (let samples = 0; samples < sp; samples++) {
        // Initialize an array with a single 0 element
        let r = [0]
        // Generate a random number between 0 and 50
        let n = _.random(0, 50)
        while (r.length < 1024) {
            // Decrement n and ensure it's non-negative
            n = n - 1
            n = n < 0 ? 0 : n
            // Generate another random number between 10 and 20
            let n2 = _.random(10, 20)

            // Add 1s and 0s to the array based on the random numbers
            for (let x = 0; x < n; x++) {

                r.push(1)
            }

            for (let x = 0; x < n2; x++) {
                r.push(0)
            }

        }
        // Divide the array into chunks of 4 elements
        s.push(_.chunk(r.slice(0, 1024), 4))
        r = []

    }

    return s

}

// Generate the training and testing datasets
let data = buildDataset(2048)
let dataTest = buildDataset(4)

// Create arrays for training and testing data
let xsTrain = []
let ysTrain = []
let xsTest = []
let ysTest = []

// Populate the training and testing arrays with data from the generated datasets
for (let s = 0; s < data.length; s++) {
    for (let k = 0; k < data[s].length; k++) {
        let pool = data[s].slice(k, 64 + k)
        if (pool.length == 64) {
            xsTrain.push(_.take(pool, 32))
            ysTrain.push(_.takeRight(pool, 32))
        }
    }
}
for (let s = 0; s < dataTest.length; s++) {
    for (let k = 0; k < dataTest[s].length; k++) {
        let pool = dataTest[s].slice(k, 64 + k)
        if (pool.length == 64) {
            xsTest.push(_.take(pool, 32))
            ysTest.push(_.takeRight(pool, 32))
        }
    }
}

// Define the input layer of the neural network
let input = tf.layers.input({
    shape: [32, 4]
})

// Add convolutional and reshaping layers to the network
let x = tf.layers.reshape({ targetShape: [2, 4, 4, 4] }).apply(input)
x = tf.layers.conv3d({ filters: 32, kernelSize: 5, padding: 'same', activation: "tanh" }).apply(x)
x = tf.layers.conv3d({ filters: 32, kernelSize: 5, padding: 'same', activation: "tanh" }).apply(x)
x = tf.layers.conv3d({ filters: 32, kernelSize: 5, padding: 'same', activation: "tanh" }).apply(x)
x = tf.layers.conv3d({ filters: 32, kernelSize: 1, padding: 'same', activation: "tanh" }).apply(x)
x = tf.layers.reshape({ targetShape: [32, 4,8] }).apply(x)

// Add a dense layer with softmax activation to the network
let x1 = tf.layers.dense({
    units: 2,
    activation: "softmax"
}).apply(x)

// Define the neural network model
let model = tf.model({
    inputs: [input],
    outputs: [x1]
})

// Compile the model with a mean squared error loss function,
// an optimizer, and a metric for evaluation
model.compile({
    loss: 'sparseCategoricalCrossentropy',
    metrics: ['acc'],
    optimizer: tf.train.adam(.0001)

})

// Define a function to generate a data iterator for training
function makeIterator() {
    let n = 0
    let idx=_.shuffle(new Array(ysTrain.length).fill(0).map((a,i)=>i))
    const iterator = {
        next: x => tf.tidy(() => {
            let result;

            let px = []
            let py = []
          
            n++
            for (let k = 0; k < 128; k++) {

            
                let u=idx[(n + k) % xsTrain.length]
                let xs = xsTrain[u]
                let ys = ysTrain[u]

                px.push(xs)
                py.push(ys)

            }

            let tx1 = tf.tensor(px)

            let ty1 = tf.tensor(py).expandDims(-1)
            result = {
                value: {
                    xs: tx1,
                    ys: ty1
                },
                done: false
            };

            return result;

        })
    }

    return iterator;
}

// Create a TensorFlow.js dataset from the data iterator
const ds = tf.data.generator(makeIterator);

// Train the model using the dataset
model.fitDataset(ds, {
    batchesPerEpoch: xsTest.length,
    epochs: 100,
    verbose:1,
    callbacks:{
        onEpochEnd(){
            // Print the prediction accuracy of the model on a random subset of the testing set
            // every epoch
            for(let x=0;x<5;x++){
let k=_.random(0,xsTest.length-1)
            let res=model.predict(tf.tensor(xsTest)).argMax(-1).arraySync()
            console.log('sample',x)
            console.log('input  :',xsTest.map(s=>s.map(s=>s.join('')).join(''))[k].split('').map(s=>s=='1'?colors.bgBlue(' '):colors.bgWhite(' ')).join(''))
            console.log('target :',ysTest.map(s=>s.map(s=>s.join('')).join(''))[k].split('').map(s=>s=='1'?colors.bgBlue(' '):colors.bgWhite(' ')).join(''))
            console.log('predict:',res.map(s=>s.map(s=>s.join('')).join(''))[k].split('').map(s=>s=='1'?colors.bgRed(' '):colors.bgWhite(' ')).join(''))
            }
        }
    }
})