9cbd2d98f7f2230fd0eb1248c361728805e2c784,PyPi/algorithms/td.py,WeightedQLearning,_next_q,#WeightedQLearning#Any#,184

Before Change

        
        means = np.zeros((1, self.mdp_info["action_space"].n))
        sigmas = np.zeros(means.shape)
        actions = self.mdp_info["action_space"].values
        for i, a in enumerate(actions):
            sa_n = [next_state, np.array([a])]
            sa_n_idx = tuple(np.concatenate((next_state, np.array([a])),
                                            axis=1).astype(np.int).ravel())
            means[0, i] = self.approximator.predict(sa_n)
            sigmas[0, i] = self._sigma[sa_n_idx]

        if self.sampling:
            samples = np.random.normal(np.repeat(means, self.precision, 0),
                                       np.repeat(sigmas, self.precision, 0))
            max_idx = np.argmax(samples, axis=1)
            max_idx, max_count = np.unique(max_idx, return_counts=True)
            count = np.zeros(actions.shape[0])
            count[max_idx] = max_count

            w = count / self.precision
        else:
            raise NotImplementedError

        sa = [np.repeat(next_state, actions.shape[0], axis=0), actions]
        W = np.dot(w, self.approximator.predict(sa))

        return W

After Change

        else:
            raise NotImplementedError

        return np.dot(w, means.T)[0]


class SpeedyQLearning(TD):

Instances