Which Deep Learning Frameworks Are Most in Demand? 2018 Power Scores

towardsdatascience.com

14 points·jeffhale··8 comments

Isn't keras just a high level API for various NN backends such as tensorflow and theano? Counting it as its own category seems silly.

Thank you for your message. I thought about that issue a bit as I was writing this. It isn't a perfect comparison. However, I think comparing both the higher and lower level APIs together in terms of usage and popularity does provide value when trying to measure usage and interest.

Yeah, listing Keras and FastAI in the same category as other original frameworks seems wrong to me.

On the other hand you should have absolutely included Caffe, as it is one of the major frameworks, and while there's a plan to merge it with Pytorch, it's hasn't happened yet (expected in Pytorch 1.0), and meanwhile if you want to use Caffe2 it's a very different tool from Pytorch.

Also, should have included Chainer.

Thank you so much for your message. I'm going to add Caffe because of it. I misread its status earlier. I might add Caffe2 and Chainer, but they score very low.

How is Theano (2007) “younger” than TensorFlow (2015). Even if you trace TensorFlow to DistBelief, it is still 2011.

Thank you for your comment. I state Theano is from 2011. Did I mistype somewhere and say it is younger than TensorFlow?

Improved Google Search and Arxiv metrics at Francois Chollet's suggestion.

Updated to include Caffe, DL4J, Caffe2, and Chainer. Thank you for the comments!

Reply on news.ycombinator.com