Reputation: 81
How to make new tensorflow library ? For e.g. I want to write some tensorflow code, matrix multiplication:
# File custom_matmul.py
import tensorflow as tf
matrixA = ...
matrixB = ...
matrixC = tf.matmul(...)
And in another python file, I want to do something like this:
import tensorflow as tf
tf.custom_matmul...
Or make library to be part of library layers
tf.layers.custom_matmul
Much more simpler, I want to make my own library and integrate it to the standard tensorflow library/framework.
Upvotes: 0
Views: 42
Reputation: 1155
I would recommend making your own module that imports and extends tensorflow the way you want rather than changing the source. For example
import tensorflow as tf
import mytf
Then you could do
mytf.layers.custom_matmul
you could even, within mytf, expose your own attributes to tensorflow. I wouldn't recommend this because it would be confusing to anyone reading your code (including possibly yourself in the future), but you could
tf.layers.custom_matmul = MyCustomMatMul()
you could then use tf.layers.custom_matmul as described in the question. Again, not recommended.
Upvotes: 1