Implementation is a breeze

What are the best aspects of this product?

I appreciate the high level abstraction of a lot of the functionalities. For instance, backpropagating over my entire network only requires me to run the optimizer.optimize() command.

What aspects are problematic or could work better?

It can be very hard to see variable values unless we run a TF.session. This, in turn, makes it very difficult to debug.

What specific problems in your company were solved by this product?

This tool has been essential to use when testing new algorithms as well as in designing neural networks.

Are you a current user of this product?

No