karpathy / convnetjs

Deep Learning in Javascript. Train Convolutional Neural Networks (or ordinary ones) in your browser.
MIT License
10.9k stars 2.04k forks source link

Question: TF & convnetjs for a JS tsumego solver #89

Open d180cf opened 7 years ago

d180cf commented 7 years ago

Hi Andrew,

Do you think it's the best approach to construct and train a simple DCNN in tensorflow and then run the trained NN with convnetjs in a JS app?

Basically, I'm writing a tsumego solver in JS and recently I've found a paper that describes promising results in application of NNs to tsumego solving. I want to replicate that NN, train on my data, make a couple improvements and use in my solver.

TF is supposed to be very fast as it uses C++ back-end with state-of-the-art (I hope) matrix-related and other algorithms. However somewhere on stackoverflow people claimed that convnetjs is way faster despite it's in JS. I'm a newbie in machine learning (still completing the TF tutorial), hence asking.

Regards.