Function softmax

rl3/flappy2envs.py:64–67

(a)

K = 2

def softmax(a):
  c = np.max(a, axis=1, keepdims=True)
  e = np.exp(a - c)
  return e / e.sum(axis=-1, keepdims=True)

def relu(x):
  return x * (x > 0)