keras 2.1.2 theano 0.9 gpu error about capsule HOT 14 CLOSED

oh, sorry. I has not any experience about theano.
it seems that T.nnet.softmax only support 1-d or 2-d tensor. but keras.io do not show it.

I will try to fix it but I am sure about it.

from capsule.

Thanks a lot.

from capsule.

now I change the code, and I build the capsule model successfully with Keras + theano, but error while training with model.fit. I has debuged long time but can not find the reason.

If you can work out, please tell me.

from capsule.

@bojone
Maybe it's about the wrong output shape of the K.batch_dot(), which is a Runtime Error

        for i in range(self.routings):
            c = softmax(b, 1)
            outputs = self.activation(K.batch_dot(c, u_hat_vecs, [2, 2]))   # outputs.shape = [None, num_capsule, num_capsule, dim_capsule]
            if i < self.routings - 1:
                b = K.batch_dot(outputs, u_hat_vecs, [2, 3])

the outputs's shape does not match the output of compute_output_shape.

Best,
Yiming

from capsule.

@ymcui
c.shape is [None, num_capsule, input_num_capsule]
u_hat_vecs.shape = [None, num_capsule, input_num_capsule, dim_capsule]
batch_dot of them has shape of [None, num_capsule, dim_capsule]
why do you think of [None, num_capsule, num_capsule, dim_capsule] ?

from capsule.

@bojone Maybe you can have a try with toy tensor.
The result should be the shape that I have posted.
Your error results from the runtime error on shape.
Just a quick fix: https://github.com/ymcui/Capsule/blob/master/Capsule_Keras.py

from capsule.

@ymcui you mean batch_dot has a different performance with theano backend?

from capsule.

@ymcui you are right. I got it. thanks very much.

from capsule.

You're welcome :)

from capsule.

when I run "python capsule_test.py",there is some error as follows:
Traceback (most recent call last):
File "capsule_test.py", line 82, in
cnn = Reshape((-1, 128))(cnn)
File "/usr/local/lib/python2.7/dist-packages/Keras-2.0.2-py2.7.egg/keras/engine/topology.py", line 554, in call
output = self.call(inputs, **kwargs)
File "/usr/local/lib/python2.7/dist-packages/Keras-2.0.2-py2.7.egg/keras/layers/core.py", line 402, in call
target_shape = self.compute_output_shape(input_shape)[1:]
File "/usr/local/lib/python2.7/dist-packages/Keras-2.0.2-py2.7.egg/keras/layers/core.py", line 385, in compute_output_shape
input_shape[1:], self.target_shape)
File "/usr/local/lib/python2.7/dist-packages/Keras-2.0.2-py2.7.egg/keras/layers/core.py", line 373, in _fix_unknown_dimension
original = np.prod(input_shape, dtype=int)
File "/usr/local/lib/python2.7/dist-packages/numpy/core/fromnumeric.py", line 2518, in prod
out=out, **kwargs)
File "/usr/local/lib/python2.7/dist-packages/numpy/core/_methods.py", line 35, in _prod
return umr_prod(a, axis, dtype, out, keepdims)
TypeError: long() argument must be a string or a number, not 'NoneType'

I update keras,but the error remains,please help me

from capsule.

@wuzy361 which backend？

from capsule.

using tensorflow backend.

when I run "python capsule_test.py",there is some error as follows:
Traceback (most recent call last):
File "capsule_test.py", line 83, in
capsule = Capsule(10, 16, 3, True)(cnn)
File "/usr/local/lib/python2.7/dist-packages/keras/engine/topology.py", line 619, in call
output = self.call(inputs, **kwargs)
File "/home/zhangjunwang/Capsule1/Capsule_Keras.py", line 71, in call
o = K.l2_normalize(o, -1)
File "/usr/local/lib/python2.7/dist-packages/keras/backend/tensorflow_backend.py", line 3151, in l2_normalize
return tf.nn.l2_normalize(x, axis=axis)
TypeError: l2_normalize() got an unexpected keyword argument 'axis'

please help me!Thanks

from capsule.

@zhangjunwang
your keras and tensorflow must be updated to lastest version.
the problem is resulted from that your tensorflow version is lower than the keras needed.

from capsule.

@bojone
Thanks,I will try it.

from capsule.