Hi Liran,

You can add million of variations to each model and there are endless of options that will help you to improve your classification, but you need to remember that this is just a short tutorial that helps to understand the basics of PyTorch for someone that uses it for the first time and the main purpose is to work with PyTorch libraries and not to learn ML.. Later you can take my tutorial and change things as you like.

This tutorial also follows after a research that was written in TensorFlow Neural Sentiment Analyzer for Modern Hebrew. All the metrics and methods and architectures were taken from there, so I really recommend you to read it.

For model.history you can build a graph and keep the history of the metrics in an array, when I’ll have time I will show you how.

I guess that you have red my NNI tutorial, over there you can find the link to my git that explains how to add it to your code.

For GPUs use you’re right and model.to(device) should be added but should not affect of running if will be omitted.