Cannot convert a symbolic Tensor in theano.scan
IlyesAbdelhamid opened this issue · comments
Dear all,
I am trying to run the following command line:
dst, _ = theano.scan(lambda z_i: T.sqrt(((z_i - self.mu) ** 2).sum(axis=1)), sequences=z)
self.mu is declared as it follows: self.mu = K.variable(mu,dtype='float32'), with K calling the backend module from keras and T the theano.tensor module.
However it comes with this error message:
NotImplementedError: Cannot convert a symbolic Tensor (Relu_199:0) to a numpy array.
I use keras version 1.0.7, theano version 0.8.2 and tensorflow version 1.15 and python version 3.6.9.
The tensor to be looped over should be provided to scan using the sequence keyword argument as per the documentation of theano.scan. So to make sure, I am passing the required type, I print(z) right before the aforementioned command line.
Here is the outcome of print(z): Tensor("Relu_199:0", shape=(?, 10), dtype=float32)
I tried to force cast z into Tensor object (although it appears to be a tensor as per print(z) ), but in vain.
I appreciate your help in this matter.
Thank you!
Sincerely,
Ilyes