Week 4 Assignmen 2 UNQ_C5: AttributeError: 'list' object has no attribute 'get_shape'

Course Q&A Deep Learning Specialization Convolutional Neural Networks
1

When running the cell after the exercice “UNQ_C5 # GRADED FUNCTION: train_step” in Assignment 2 of week 4 I get the following error:

    <ipython-input-100-e9e1eb98ee23>:19 train_step  *
        J_style = compute_layer_style_cost(a_S, a_G)
    <ipython-input-83-afd748e746db>:16 compute_layer_style_cost  *
        _, n_H, n_W, n_C = a_G.get_shape().as_list()

    AttributeError: 'list' object has no attribute 'get_shape'

So the problem seems to be in how a_G is defined, which I believe there is just one way to implement it by following the instructions “Compute the encoding of the generated image using vgg_model_outputs. Assing the result to a_G”.

A hint will be very much appreciated.

2

First of all, please see this thread.

There are 6 outputs in a list. That’s a\_G. For the style cost calculation, we need to handle 5 outputs in a list. On the other hand, “compute_layer_style_cost()” is for one single layer (output) which is unpacked from a list. (In this sense, if you pass this list to “compute_layer_style_cost()”, then, the above error should occur.)
In stead, you need to call other function which actually calls “compute_layer_style_cost()” with unpacking a list.

Hope this helps.

1 Like
3

Thank for the quick answer. I see now that I was using the wrong funcion and the link was very useful to understand where this list of outputs comes from.

However after rebooting and rerunning I still get an error

TypeError: in user code:

    <ipython-input-26-bf53beb222ea>:19 train_step  *
        J_style = compute_style_cost(a_S, a_G)
    <ipython-input-14-d70c9653228b>:28 compute_style_cost  *
        J_style_layer = compute_layer_style_cost(a_S[i], a_G[i])
    <ipython-input-9-afd748e746db>:19 compute_layer_style_cost  *
        a_S = tf.transpose(tf.reshape(a_S,[n_H * n_W, n_C], [1,0]))
    /usr/local/lib/python3.6/dist-packages/tensorflow/python/util/dispatch.py:201 wrapper  **
        return target(*args, **kwargs)
    /usr/local/lib/python3.6/dist-packages/tensorflow/python/ops/array_ops.py:195 reshape
        result = gen_array_ops.reshape(tensor, shape, name)
    /usr/local/lib/python3.6/dist-packages/tensorflow/python/ops/gen_array_ops.py:8234 reshape
        "Reshape", tensor=tensor, shape=shape, name=name)
    /usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/op_def_library.py:353 _apply_op_helper
        with g.as_default(), ops.name_scope(name) as scope:
    /usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/ops.py:6492 __enter__
        return self._name_scope.__enter__()
    /usr/lib/python3.6/contextlib.py:81 __enter__
        return next(self.gen)
    /usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/ops.py:4189 name_scope
        if not _VALID_OP_NAME_REGEX.match(name):

    TypeError: expected string or bytes-like object
4

Check the location of “close” parenthesis for tf.reshape(). It needs first two. Third one, [1,0], should be for tf.transpose() .

5

Thanks again for the very fast response. It worked. It is funny that I passed the test and got the right value for the “UNQ_C3, GRADED FUNCTION: compute_layer_style_cost” even with the issues with the parenthesis.

6

As you are using “Positional Argument”, an interpretation is what Tensorflow assumes. The third position for tf.Reshape is name which expects “string”. We do not know how [1,0] is handled in tf.Reshape, but, one big difference is “complete_layer_style_cost” is called from “train_step” which is inside “@tf.function”. This changes an execution mode from “Eager execution” to “Graph execution”, which runs faster. We can easily imagine that this additional parameter, [1,0], caused an error in a graph computation. You can put @tf.function in front of “def compute_layer_style_cost(a_S, a_G):” to see what will happen. I believe the same error occurs even by a local test. :wink: