Yeah, that was my very first one. I have since started with vectorized implementations instead of that clumsy way I did it the first time. However I found vectorization a bit tricky at first.
I'm actually planning to try and learn convolutional nets soon, and will try to setup a small study group online (basically just a repo with materials and a Slack/Gitter chat room), where people can help each other with learning it. Do you think that would help you to learn it?
Interesting how you haven't used any machine learning framework :) Did you do it purposely to get a better understanding of neural nets? Are you going to move to frameworks to speed up your iteration speed? Or are you more interested in maybe refactoring your code into a reusable framework? :)
Yeah, I wanted to understand how they worked, so I stayed away from any framework.
I have since moved over to using Theano, Lasagne & nolearn. However, I almost find it just as hard, because it's so easy to step wrong and get errors, and so hard to debug. In addition, Lasagne & nolearn doesn't exactly have great documentation of tutorials available (they have some, but I like to try out many tutorials to get perspective and get a proper understanding).
I'm far from good enough to turn my implementations into frameworks :p
I'm actually planning to try and learn convolutional nets soon, and will try to setup a small study group online (basically just a repo with materials and a Slack/Gitter chat room), where people can help each other with learning it. Do you think that would help you to learn it?