3 ms·
While we are at it, which framework is the easiest to tweak at the low-level, e.g. create modified LSTMs etc ?
by zerostar07 8y ago
While we are at it, which framework is the easiest to tweak at the low-level, e.g. create modified LSTMs etc ?
- laingc 8y agoPytorch, by a mile. If I had to summarise the frameworks in a few words, they would be: Keras for speed, Tensorflow for production, Pytorch for research.
- mlthoughts2018 8y agoI would say keras or mxnet for speed and production. PyTorch for research. By this point there are hardly any cases when it’s worth it to descend to lower TensorFlow levels.
- pdyck 8y agoKeras only offers standard layers but you can implement your own LSTM layer and use it with Keras. This way you can take advantage of all the other features