Hi everyone,
Here is a short clip of a light-weight project I made using OpenAI's gym. If you've ever tried to balance an umbrella on your palm, you'll get the essence of it.
Over here, I've implemented a deep neural net (with only numpy) with 2 hidden layers which learns by itself how to balance a cart-pole system from scratch with zero supervision.
The neural net has 2 hidden layers with 10 neurons each with ReLU activations and a softmax output.
Find the training code here - https://goo.gl/1FxXm6
Neural Net class - https://goo.gl/twvu9v
Follow me on GitHub at https://github.com/ironhide23586
Thanks!