bbd50fcbc6d67b7acba0a794b1cdb47ccf5731e2,ml/rl/models/actor.py,DirichletFullyConnectedActor,_get_concentration,#DirichletFullyConnectedActor#Any#,196

Before Change

        Get concentration of distribution.
        https://stats.stackexchange.com/questions/244917/what-exactly-is-the-alpha-in-the-dirichlet-distribution
        
        return self.fc(state.float_features)

    def get_log_prob(self, state, action):
        with torch.no_grad():
            concentration = self._get_concentration(state)

After Change

        Get concentration of distribution.
        https://stats.stackexchange.com/questions/244917/what-exactly-is-the-alpha-in-the-dirichlet-distribution
        
        return self.fc(state.float_features) + self.EPSILON

    def get_log_prob(self, state, action):
        with torch.no_grad():
            concentration = self._get_concentration(state)

Instances