George Hotz 30 Aug 2018 programming live-stream

#!/usr/bin/env python3
# goal is to transform uniform(0,1) into N(-1,1)

import matplotlib.pyplot as plt
import random
import numpy as np

from keras import optimizers
from keras.models import Sequential
from keras.layers import Dense, Activation

# generator
g = Sequential([
  Dense(3, input_shape=(1,)),
  Activation('elu'),
  Dense(3),
  Activation('elu'),
  Dense(1),
])

# discriminator
# output 1 for real, 0 for fake
# nobody likes fake people
d = Sequential([
  Dense(3, input_shape=(1,)),
  Activation('elu'),
  Dense(3),
  Activation('elu'),
  Dense(1),
  Activation('sigmoid'),
])

#sgd = optimizers.SGD(lr=0.01, momentum=0.0, nesterov=True)
sgd = optimizers.SGD()

g.compile(optimizer=sgd, loss='mse')
d.compile(optimizer=sgd, loss='mse')

for loop in range(10):
  print("******* loop %d *******" % loop)
  # train discriminator
  d.trainable = True
  X, Y = [], []

  Z = np.array([random.random() for x in range(16384)])
  samples_fake = g.predict(Z).flatten()
  samples_real = np.random.normal(-1.0, 1.0, 16384)

  plt.hist(samples_real, bins=np.arange(-3,3,0.1))
  plt.hist(samples_fake, bins=np.arange(-3,3,0.1))
  plt.show()

  X = list(samples_fake) + list(samples_real)
  Y = [0]*samples_fake.shape[0] + [1]*samples_real.shape[0]

  # shuffle together
  XY = list(zip(X,Y))
  random.shuffle(XY)
  X,Y = zip(*XY)
  d.fit(np.array(X), np.array(Y), epochs=3)

  # train generator
  d.trainable = False
  Z = np.array([random.random() for x in range(16384)])
  gd = Sequential([g, d])
  gd.compile(optimizer=sgd, loss='mse')
  gd.fit(Z, np.array([1]*Z.shape[0]), epochs=3)