CODEWITHSUNDEEP
node.jstensorflowmachine-learningartificial-intelligencetensorflow.js
clota974
clota974

Reputation: 505

Beginner : Not understanding shape [,1] on Tensorflow.js

I want to make a model to convert Celsius to Fahrenheit with Tensorflow.js (with Node.js).

However, I am not understanding what shapes to use.

I have tried different input_shape such as [1], [1,20] and finally set it to[20] I also have tried different tensor shape for the Celsius & Fahrenheit arrays such as tensor(celsius), tensor([celsius]).

Here is the code


var model = tf.sequential()
model.add(tf.layers.dense({inputShape:[20], units: 1}))

async function trainModel(model, inputs, labels) {
    // Prepare the model for training.  
    model.compile({
      optimizer: tf.train.adam(),
      loss: tf.losses.meanSquaredError,
      metrics: ['mse'],
    });

    const batchSize = 28;
    const epochs = 500;

    return await model.fit(inputs, labels, {
      epochs,
      shuffle: true,
    //   callbacks: tfvis.show.fitCallbacks(
    //     { name: 'Training Performance' },
    //     ['loss', 'mse'], 
    //     { height: 200, callbacks: ['onEpochEnd'] }
    //   )
    });
  }

c = tf.tensor([celsius]) // celsius = [1,2,3,4,...]
console.log(c.shape) // ==> [1,20]

f = tf.tensor([fahrenheit])
console.log(f.shape) // ==> [1,20]

trainModel(model, c, f)

Moreover, in the Python tutorial input_shape is [1] . With Node.js, only [20] seems to work.

The shape of inputs is [1,20] and it is correct.

Shape of labels is [1,20] as well but triggers following error :

Debugger says :

Error when checking target: expected dense_Dense1 to have shape [,1], but got array with shape [1,20].

- EDIT

Furthermore, when i try input_shape: [1,20], it gives me : 

expected dense_Dense1_input to have 3 dimension(s). but got array with shape 1,20

-

I expect the model to train by associating C° values to F° values.

Thank you

Upvotes: 1

Views: 514

Answers (1)

edkeveked
edkeveked

Reputation: 18381

The error is clear:

{inputShape:[20], units: 1}

The model contains a single layer. inputShape:[20] which means batchInputShape that is [null, 20] will be the shape of the first layer. Likewise, units: 1 indicates that the last layer will have the shape [null, 1].

The features used have the shape [1, 20] matching therefore the batchInputShape of the model. However, that's not the case for the labels that have the shape [1, 20]. It has to have the shape [1, 1] therefore throwing the error:

expected dense_Dense1 to have shape [,1], but got array with shape [1,20]

The units size of the model has to be changed to reflect the labels shape. 

{inputShape:[20], units: 20}

Upvotes: 1

Related Questions