Simple Recurrent Neural Network

Overview

Create the Model

const model = tf.sequential({
    layers:[
      tf.layers.simpleRNN({
        units:3,
        inputShape:[2,1]
      })
    ]
  });
				

An alternative approach is to create the model and then add the layers.

const meodel = tf.sequential();
model.add(tf.layers.simpleRNN({
	units:3,
	inputShape:[2,1]
}));
				

Prediction

The following script implements a simple RNN and then runs it on a sample input. No training is done, so the output will be random.

//load tensor flow library
await $src('https://cdn.jsdelivr.net/npm/@tensorflow/tfjs@latest');
let ut = await import('/lib/tensor-flow/util/v1.0.0/util.mjs');

const model = tf.sequential({
    layers:[
      tf.layers.simpleRNN({
        units:3,
        inputShape:[2,1]
      })
    ]
  });

let input = tf.tensor([[[1],[2]]]);
let output = model.predict(input);
let matrix = await ut.toArray(output);

				

Try it!

Training

The following script implements a simple RNN and then runs it on a sample input. No training is done, so the output will be random.

//load tensor flow library
await $src('https://cdn.jsdelivr.net/npm/@tensorflow/tfjs@latest');
let ut = await import('/lib/tensor-flow/util/v1.0.0/util.mjs');

const model = tf.sequential({
    layers:[
      tf.layers.simpleRNN({
        units:3,
        inputShape:[2,1]
      })
    ]
  });
  
model.compile({
  loss:tf.losses.meanSquaredError,
  optimizer:tf.train.adam(0.1)
});

let X = [
    [[1],[2]],
    [[2],[2]]
];
let xs = tf.tensor(X);

let y= [
    [1,1,1],
    [1,3,1]
];
let ys = tf.tensor(y);

await model.fit(xs,ys,{
  batchSize:2,
  shuffle:true,
  epochs:10
});
				

Try it!