In which areas doesn't PyTorch yet provide good solutions?

[nbro]

I like PyTorch, but I found that it still lacks some if not many useful and reliable tools/libraries/solutions, in certain areas. The aim of this post is to encourage people to list the areas in which PyTorch still doesn’t provide reliable, complete, elegant and “production-ready” solutions.

Here are a few of them

• k-fold cross-validation (see e.g. What is the best way to apply k-fold cross validation in CNN?)
• Hyper-parameter optimization (see e.g. What is the best way to perform hyper parameter search in PyTorch?)
• Visualization tools (https://nbro.github.io/blogging/2019/01/06/an-example-of-how-to-use-visualdl-with-pytorch/)

[Jakub_Czakon]

Hi @nbro,

Jakub from neptune.ml here.
I have recently developed a utility that helps you run pretty much any hyperparameter optimization library on pretty much any python script. You can check it out here with examples in scikit-optimize and hyperopt.

Before you ask, Neptune is open and free for non-organizations.
Read more about it in the docs page to get a better view.