Tensorflow.js tf.Sequential class .evaluateDataset() Method

Tensorflow.js is an open-source library that is developed by Google for running machine learning models as well as deep learning neural networks in the browser or node environment.

The .evaluateDataset() function is used to evaluate the stated sequential model by means of a stated dataset object.

Note: This method differs from evaluate(), as it is asynchronous.

Syntax:

evaluateDataset(dataset, args?)

Parameters:  

• dataset: The stated dataset object. Whose iterator() method is awaited in order to produce a dataset iterator object, its next() method is awaited to generate data batches for evaluation. Moreover, the return value of the next() call must hold a boolean _done field as well as a _value field. Where, the value field is likely to be an array of two tf.Tensors or else an array of two nested tf.Tensor builds. The first case is for models with exactly one input and one output (e.g.. a sequential model). The latter case is in favor of the models along with several inputs and/or several outputs. Out of two elements in the stated array, the first one is the input feature(s) and the second one is the output target(s). It is of type tf.data.Dataset.
• args: The stated configuration object in favor of the dataset based evaluation. It is optional and is of type object. The following arguments come under it:
  1. batches: The stated number of batches which is to be dragged from the given dataset object before terminating the evaluation. It is of type number.
  2. verbose: The stated verbosity mode. It is of type ModelLoggingVerbosity.

Return Value: It returns promise of tf.Scalar or tf.Scalar[].

Example 1:

Javascript

// Importing the tensorflow.js library
import * as tf from "@tensorflow/tfjs"

// Defining an array x
const Array_x = [
   [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],
];

// Defining an array y
const Array_y = [1, 1, 1, 1];

// Defining dataset of x
const Dataset_x = tf.data.array(Array_x);

// Defining dataset of y
const Dataset_y = tf.data.array(Array_y);

// Defining dataset of xy using zip method
const Dataset_xy = tf.data.zip({xs: Dataset_x, ys: Dataset_y})
     .batch(5)
     .shuffle(3);

// Training Model
const gfg = tf.sequential();

// Adding layer to model
const layer = tf.layers.dense({units:1,
               inputShape : [8]});
   gfg.add(layer);

// Compiling our model
const config = {optimizer:'sgd',
              loss:'meanSquaredError'};
  gfg.compile(config);

// Calling evaluateDataset() method
const res = await gfg.evaluateDataset(Dataset_xy,
tf.ones([7, 10]), tf.ones([7, 1]), {
   batchSize: 5,
});

// Printing output
res.print();

Output:

Tensor
    2.9478049278259277

Example 2:

Javascript

// Importing the tensorflow.js library
import * as tf from "@tensorflow/tfjs"

// Defining dataset of xy using zip method
const Dataset_xy = tf.data.zip({
    xs: tf.data.array([[1, 0, 1, 2, 1]]),
    ys: tf.data.array([1, 2, 1, 3])}).batch(8);

// Training Model
const gfg = tf.sequential();

// Adding layer to model
const layer = tf.layers.dense({units:1,
    inputShape : [5], activation: 'sigmoid'});

gfg.add(layer);

// Compiling our model
const config = {optimizer:'sgd',
              loss:'meanSquaredError'};

gfg.compile(config);

// Calling evaluateDataset() method
const res = await gfg.evaluateDataset(
    Dataset_xy, tf.truncatedNormal([7, 10]),
    tf.randomNormal([7, 1]),
    {batchSize: 2, steps: 1});

// Printing output
console.log(JSON.stringify(res));

Output:

{"kept":false,"isDisposedInternal":false,"shape":[],
"dtype":"float32","size":1,"strides":[],"dataId":{"id":7097},
"id":4731,"rankType":"0","scopeId":4352}

Reference: https://js.tensorflow.org/api/latest/#tf.Sequential.evaluateDataset