greydanus

'''Solves Pong with Policy Gradients in Tensorflow.'''
# written October 2016 by Sam Greydanus
# inspired by karpathy's gist.github.com/karpathy/a4166c7fe253700972fcbc77e4ea32c5
import numpy as np
import gym
import tensorflow as tf

# hyperparameters
n_obs = 80 * 80           # dimensionality of observations
h = 200                   # number of hidden layer neurons

greydanus

"Dynamic plotting in matplotlib. Copy and paste into a Jupyter notebook."
# written October 2016 by Sam Greydanus
%matplotlib notebook
import matplotlib.pyplot as plt
import numpy as np
import time

def plt_dynamic(x, y, ax, colors=['b']):
    for color in colors:
        ax.plot(x, y, color)

greydanus

'''Solution to the Cartpole problem using Policy Gradients in Tensorflow.'''
# written October 2016 by Sam Greydanus
# inspired by gist.github.com/karpathy/a4166c7fe253700972fcbc77e4ea32c5
import numpy as np
import gym
import tensorflow as tf

# hyperparameters
n_obs = 4              # dimensionality of observations
h = 128                # hidden layer neurons

greydanus

""" Trains an MNIST classifier using Synthetic Gradients. See Google DeepMind paper @ arxiv.org/abs/1608.05343. """
import numpy as np
import matplotlib.pyplot as plt
import matplotlib.cm as cm
from tensorflow.examples.tutorials.mnist import input_data # just use tensorflow's mnist api
mnist = input_data.read_data_sets('MNIST_data', one_hot=False)

# hyperparameters
global_step = 0
batch_size = 10