How to efficiently process a batch of terms in one go for Tree Neural Networks?

[brando90]

this seems like an important point so I wanted to make a discussion. I was wondering if the code that batches ASTs has really been tested to see if it's faster?

#16

[yangky11]

I may have measured the speed and found it to be faster when I wrote the code. I'm not sure though at this point.

[brando90]

I am curious, is there a reason you didn't use these two options for batching:

• https://developer.nvidia.com/blog/recursive-neural-networks-pytorch/
• https://github.com/unbounce/pytorch-tree-lstm

?

[yangky11]

I was unaware of these two repos. Also, since batching for TreeLSTM is not too complex, it was not worth introducing another 3rd-party dependency.