Doing some data crunching. Graph: daily number of new StackOverflow questions that either contain one of {"keras", "tensorflow"} in their tags or titles, or one of {"torch", "pytorch", "torchvision"}. Past ~6 months. 
75 questions per day on average -- you could employ a few people full time just answering them
Daily PyPI downloads (EMA). Labeled this time.
Now here's a really fascinating one. Weekly number of research papers indexed on arXiv specifically, and on all of Google Scholar (which includes arXiv), that mention TensorFlow, Keras, or PyTorch.
This tells us two things:
1) The pure ML research community (posts on arXiv) is evenly split between TF and PyTorch
2) The broader scientific community interested in applying ML (doesn't post on arXiv; often doesn't even publish preprints), which is ~20x larger, uses TF/Keras
1) The pure ML research community (posts on arXiv) is evenly split between TF and PyTorch
2) The broader scientific community interested in applying ML (doesn't post on arXiv; often doesn't even publish preprints), which is ~20x larger, uses TF/Keras
