Open quantuminformation opened 6 years ago
I have done some work with both. I'd love to work with Synaptic only.
Synaptic is easier to work with (clean, easy to understand API and proper documentation). It's also as fast (or faster) for small or simple networks. V.2 seems to take time - but might be an excellent thing if/when the plans become reality. Neataptic is an interesting spin-off supporting neuroevolution, if that is something you'd like.
Deeplearn.js on the other hand has support for GPU, making it much, much faster for complex networks. It also has convolutional and max pooling layers - a must for working with images, IMHO. The API is complex, the documentation is out of synch with the latest version. Everything seems to be changing a lot. Not very developer friendly, I'm afraid.
Reminds me of Tensorflow vs. Keras. Keras being like Synaptic in ease-of-use. But since I need GPU and images, I started a project just before the holidays; aiming at adding a Keras/Synaptic -like API layer on top of deeplearn.js.
Thanks mate, I need to do a lot with images too. Feel free to join my discord for deeplearn.js to talk further if you like.
I have a project that needs machine learning in the browser, was wondering if someone could point me in the right direction.