What keras updates make error of this simple code?

Question

To use multi-GPU, I upgrade keras(1.2.0 --> 2.2.4) version. I don't change any of code, but there is an error. What is the reason of this ERROR?

I attached the reproducible code below. As you can see A & B have same shapes, but only testB got the dimension error.

< Keras 1.2.0 LOGS >

> pip3 install keras==1.2.0 --user
> python test.py
Using TensorFlow backend.
Done

< Keras 2.2.4 LOGS >

> pip3 install git+git://github.com/fchollet/keras.git --upgrade --user
> python test.py
Using TensorFlow backend.
test.py:26: UserWarning: Update your `Conv1D` call to the Keras 2 API: `Conv1D(128, 3, use_bias=False)`
  testA = keras.layers.Conv1D(128,3,bias=False)(A)
test.py:27: UserWarning: Update your `Conv1D` call to the Keras 2 API: `Conv1D(128, 3, use_bias=False)`
  testB = keras.layers.Conv1D(128,3,bias=False)(B)
Traceback (most recent call last):
  File "test.py", line 27, in <module>
    testB = keras.layers.Conv1D(128,3,bias=False)(B) 
  File "/home/nam/.local/lib/python3.6/site-packages/keras/engine/base_layer.py", line 457, in __call__
    output = self.call(inputs, **kwargs)
  File "/home/nam/.local/lib/python3.6/site-packages/keras/layers/convolutional.py", line 163, in call
    dilation_rate=self.dilation_rate[0])
  File "/home/nam/.local/lib/python3.6/site-packages/keras/backend/tensorflow_backend.py", line 3881, in conv1d
    data_format=tf_data_format)
  File "/home/nam/.local/lib/python3.6/site-packages/tensorflow/python/ops/nn_ops.py", line 779, in convolution
    data_format=data_format)
  File "/home/nam/.local/lib/python3.6/site-packages/tensorflow/python/ops/nn_ops.py", line 839, in __init__
    filter_shape[num_spatial_dims]))
ValueError: number of input channels does not match corresponding dimension of filter, 1 != 1603

This is the reproducible code.

import keras

class AddSingletonDepth(keras.layers.Layer):

    def call(self, x, mask=None):
        x = keras.backend.expand_dims(x, -1)  # add a dimension of the right

        if keras.backend.ndim(x) == 4:
            return keras.backend.permute_dimensions(x, (0, 3, 1, 2))
        else:
            return x

    def get_output_shape_for(self, input_shape):
        if len(input_shape) == 3:
            return input_shape[0], 1, input_shape[1], input_shape[2]
        else:
            return input_shape[0], input_shape[1], 1


data = keras.engine.Input([1603])
A=keras.backend.expand_dims(data,-1)
B=AddSingletonDepth()(data)
testA = keras.layers.Conv1D(128,3,bias=False)(A)
testB = keras.layers.Conv1D(128,3,bias=False)(B) 
print('Done')

I expect to know the reason of the Error. And I really likes to use this formation as possible, but if it's impossible could you show me how to fix the error?

Thank you.

What keras updates make error of this simple code?

0 Answers0