Week 1 - Jazz, music_inference_model, one hot encoding

Course Q&A Deep Learning Specialization Sequence Models
1

When I run the following code from step 2.D to 2.E in music_inference_model funcion:
*# Step 2.D: *

  •    # Select the next value according to "out",*

  •    # Set "x" to be the one-hot representation of the selected value*

  •    # See instructions above.*

  •    print("A")*

  •    print(x)*

  •    x = tf.math.argmax(input=x, axis=-1)*

  •    print("B")*

  •    print(x)*

  •    x = tf.one_hot(indices=x, depth=90)*

  •    # x = tf.keras.layers.Flatten()(tf.one_hot(indices=x, depth=90))*

  •    print("C")*

  •    print(x)*

  •    # Step 2.E: *

  •    # Use RepeatVector(1) to convert x into a tensor with shape=(None, 1, 90)*

  •    x = RepeatVector(n=1)(x)*

  •    print("D")*

  •    print(x)*

I get the following printings:
A
Tensor(“input_29:0”, shape=(None, 1, 90), dtype=float32)
B
Tensor(“ArgMax_465:0”, shape=(None, 1), dtype=int64)
C
Tensor(“OneHot_462:0”, shape=(None, 1, 90), dtype=float32)

You can observe how the output of the one hot encoding layer is of shape (None, 1, 90), when in the instructions it says it should be (None, 90) - consequently need the RepeatVector. But in this case it is not, and thus RepeatVector fails with following error:
ValueError: Input 0 of layer repeat_vector_309 is incompatible with the layer: expected ndim=2, found ndim=3. Full shape received: [None, 1, 90]

To overcome that issue, I flattened the output of one_hot layer doing:
x = tf.keras.layers.Flatten()(tf.one_hot(indices=x, depth=90))

This makes the code to work, however the test in comparator(inference_summary, music_inference_model_out) fails:
AttributeError: The layer "lstm has multiple inbound nodes, with different input shapes. Hence the notion of “input shape” is ill-defined for the layer. Use get_input_shape_at(node_index) instead.

Please could you help me in both making the code to work and ensuring that the comparator does not return any error, to be able to pass the assigment? Thanks!

1 Like
2

Solved issue ^ (I was passing x instead of out to tf.math.argmax

3 Likes