Ideas are never as cool on paper as they are in your head. As the second month closes, I’m frantically trying to finish a game to stay on schedule. I’m behind on both math problems and drawings. My goal is to have Regular Ordinary Shopping Time done before March 1st. It will be about Niklas of Regular Ordinary Swedish Meal Time going to the store to acquire food. Punch other customers out of the way and acquire the items. Simple. Effective. Regular.
It's good for you.
After 100 samples and 100 epochs, we see the following. Let’s try it on a bigger chunk of the dataset.
Also, anyone who has been submitting posts, I’ve sorta shuffled around passing any of them, as it’s hard to tell which are spam. I’ll incorporate recaptcha soon, but in the meanwhile, be sure to add [notabot] to your subject if you want me to accept your posts.
I made fun of PyBrain a while back. Found myself using it again, still dissatisfied but reluctantly putting stuff together. If you’ve got a simple ML problem and can’t be arsed to hash together a backprop learner, it does serve it’s purpose. I’m still peeved, however, because it’s slow as balls. I’m aware that backprop isn’t the fastest learning algorithm, but I would have expected a little bit of a speed bump. I wonder if they’re accelerating with numpy, of if they’re doing everything sequentially.
I have a 256 input visible layer, a 512 unit hidden layer, a 128 unit layer, and a 256 unit output layer. With 1000 training examples, each epoch is taking 20-30 minutes. :\