Which to learn: TensorFlow or PyTorch?

Question

[John Ma]

Hi all,

I already have some experience in ML, and I'd like to learn deep learning in a bioinformatics context, especially I've already been using some NN-based out-of-the-box algorithms in my routine analysis. Which would you think is a better choice, Tensorflow or PyTorch?

Thanks for the answer in advance.

Answer 1

[Mensur Dlakic]

Doubt that there are many people here who have extensively used both platforms, so I think you will get answers based on what people use. I never learned to program directly in either, but have used TensorFlow (TF) for many years as a back-end to Keras. Or maybe a correct way of saying it is that I have been using Keras as a front-end for TF.

I don't have any hard proof for what I am about to say. Still, intuitively PyTorch (PT) must be an excellent piece of software as it is still very competitive with TF, despite the latter being supported by Google's machinery. Keras used to support Theano as well (and maybe still does), but Theano couldn't compete with TF. Yet PT still does. Either way you go I don't think your skills would become outdated any time soon, as both platforms figure to be around for some time.

Answer 2

[Istvan Albert]

I will also mention scikit-learn as a good library for scientists

https://scikit-learn.org/stable/

Since it is built on NumPy, SciPy and Matplotlib it integrates well with doing science.

[John Ma]

I'm already familiar with standard ML as well as scikit-learn.