Slicing a tensor by using indices in Tensorflow

Question

Basically I have a 2d array and I want to do this nice numpy-like thing

noise_spec[:rows,:cols]

in Tensorflow. Here rows and cols are just two integers.

Solution 1

Indeed, TensorFlow now has better support for slicing, so you can use the exact same syntax as NumPy:

result = noise_spec[:rows, :cols]

Solution 2

found out, it's

tf.slice(noise_spec, [0,0],[rows, cols])

Comments

[sachinruk]

unfortunately u = U[idx_i,:] doesn't work on tf 1.8.0. idx_i are random indices of rows I wanted to choose. tf.nn.embedding_lookup worked.