PyTorch impressions?

Hey guys,

I’ve noticed that PyTorch is gaining traction.
Anyone has experience with PyTorch comparing to Keras?
What are the strong features of PyTorch?


The biggest reason to use PyTorch as far as I know is the fact it is dynamic. This allows powerful functionality like debugging and different feature lengths (different size sentences without padding for example).

I haven’t used it, just regurgitating what I’ve read.

We’ll be using pytorch quite a lot in part 2. Customizing RNNs in particular is far easier than in Keras.


You got me more exited for part 2 now :smiley: