3ca448728b74a7a6186c7c76f518f1362163ce0d,mushroom/policy/gaussian_policy.py,GaussianPolicy,diff_log,#GaussianPolicy#Any#Any#,29

Before Change

    def diff_log(self, state, action):
        mu, sigma = self._compute_gaussian(state, False)
        delta = action - mu
        j_mu = np.expand_dims(self._approximator.diff(state), axis=1)

        g = j_mu.dot(delta) / sigma**2

        return g

    def set_sigma(self, sigma):

After Change

        delta = action - mu

        j_mu = self._approximator.diff(state)

        if len(j_mu.shape) == 1:
            j_mu = np.expand_dims(j_mu, axis=1)

        g = .5 * j_mu.dot(inv_sigma + inv_sigma.T).dot(delta.T)

        return g

    def set_weights(self, weights):

Instances