Seems to be moving fast:
1 Like
Hi @kip, can you please expand why torchx can help with text? I want to make sure I understand to ensure it can effectively help you. 
One of the issues with text is that you need to modify it to work with tensors and that’s going to be an interesting problem to tackle!
1 Like
One of the first steps is sentiment analysis. For sure, that requires a lot of preprocessing (segmentation, stop words elimination, stemming) and then encoding into a tensor. Other than segmentation, that can be done using Elixirs concurrent processing quite effectively.
Then with the text encoded as tensors, the various approaches of model training, logistic regression, naive bayes, lapaclain smoothing, vector space models and so on can (as best I can tell) leverage NX and Torchx.
I am a noob to most of this and am using the opportunity to give some money to Coursera and doing my assignments in Python while experimenting in parallel with Nx and friends.
4 Likes
Really excited to get started with Nx. Is it too early to expect to find hexdocs for Nx? I can’t find any, so I guess I will need to generate them myself?
Are calculations in Nx still subject to the problem of floating point precision?
Does Nx have a function to shift rows? I’m interested in handling tabular data where new rows are appended/prepended and each time, the oldest row is dropped. Would this be a good use case for Nx?
1 Like
Yes.
It depends. Tensors are either unsigned integers, signed integers, or floating points.
I am not sure how this would fit the multi-dimensional array paradigm at the moment.
1 Like
