[FIXED] How to use nn.TransTransformerEncoder from pytorch

Issue

I am trying to use PyTorch's '''nn.TransformerEncoder''' module for a classification task. I have data points of varying lengths i.e I have sequences of different lengths. All sequences have one corresponding output(target which is either 0 or 1).[![enter code here][1]][1]

This image outlines my dataset

This image shows how the sequences vary in length

However the entries of each sequence are of the same length.

I want to use this dataset to train an encoder part of the Transformer to be able to predict the corresponding outputs. How can I go about doing this? And are there any examples I can check online?

Solution

It depends on how your data actually looks like and what kind of output you expect. In general I would suggest to you to use the Transformers Library from HuggingFace, they have a lot of documentation and detailed code examples that you can work on -- plus an active forum. Here is a link to their description of Encoder-Decoder Models. I hope that helps you a bit.

Answered By - Stimmot