Method sample_action

rl3/flappy2envs.py:92–99

(self, x)

    return softmax(Z.dot(self.W2) + self.b2)

  def sample_action(self, x):
    # assume input is a single state of size (D,)
    # first make it (N, D) to fit ML conventions
    X = np.atleast_2d(x)
    P = self.forward(X)
    p = P[0] # the first row
    # return np.random.choice(len(p), p=p)
    return np.argmax(p)

  def score(self, X, Y):
    P = np.argmax(self.forward(X), axis=1)