a bunch of machine learning things

This is where I'm putting my forays into neural networks and, more generally, machine learning. Most of these are tested with and made for OpenAI Gym, a set of environments made for reinforced learning in Python. These are all my work, and don't use any external ML libraries like tensorflow.

• PIDevolve.py was my first foray into machine learning, and doesn't actually involve neural nets, but a PID controller. It was built for, and only works on, problems where the goal is to keep some quantity as constant as possible, e.g. the 'cartpole' environment in OpenAI Gym (https://gym.openai.com/envs/CartPole-v1/). It tunes the coefficients in the PID function through a genetic algorithm.

• neuralnetevolve.py is a general-purpose neural net, with weights adjusted by genetic algorithm. Works in any environment, although results may vary for the more complex ones, and evolution may take quite some time. Here's it learning how to land a lunar lander in an openAI gym environment: https://www.youtube.com/watch?v=X0UEikltbZs

• backprop1.py teaches a neural network through backwards propagation, a different method of machine learning which actively encourages good behaviour and discourages bad, through a technique known as gradient descent. The way it chooses whether to encourage or discourage is pretty rudimentary, so it's rather slow and it only works for the cartpole environment and similar.

• neuralnetcrypto.py is a trading bot for crypto markets (bitcoin, ethereum, etc). It trains on historical price data, learning to notice patterns in the Relative Strength Index and Moving Average indicators. It does pretty well atm but i'll be extending it to analyze more indicators soon:

Performance if randomly choosing to buy or sell (profit ratio vs trials) : https://i.imgur.com/9ZaayLJ.png

Performance of my neural net: https://i.imgur.com/f01FRQa.png