Python ActorCritic._choose_action示例

示例#1

文件： policy_gradient_test.py 项目： bigdatasciencegroup/Microsoft-Cognitive-Recognition-Toolkit-CNTK

    def test_rollout_preprocess(self, mock_parameters):
        self._setup_parameters(mock_parameters.return_value)
        mock_parameters.return_value.preprocessing = \
            'cntk.contrib.deeprl.agent.shared.preprocessing.SlidingWindow'
        mock_parameters.return_value.preprocessing_args = '(2, "float32")'

        action_space = spaces.Discrete(2)
        observation_space = spaces.Box(0, 1, (1, ))
        sut = ActorCritic('', observation_space, action_space)

        sut._choose_action = Mock(side_effect=[(0, ''), (1, ''), (1, '')])

        sut.start(np.array([0.1], np.float32))
        sut.step(0.1, np.array([0.2], np.float32))
        sut.step(0.2, np.array([0.3], np.float32))

        self.assertEqual(sut._trajectory_rewards, [0.1, 0.2])
        self.assertEqual(sut._trajectory_actions, [0, 1, 1])
        np.testing.assert_array_equal(sut._trajectory_states, [
            np.array([[0], [0.1]], np.float32),
            np.array([[0.1], [0.2]], np.float32),
            np.array([[0.2], [0.3]], np.float32)
        ])

        sut.end(0.3, np.array([0.4], np.float32))

        self.assertEqual(sut._trajectory_rewards, [0.1, 0.2, 0.3])
        self.assertEqual(sut._trajectory_actions, [0, 1, 1])
        np.testing.assert_array_equal(sut._trajectory_states, [
            np.array([[0], [0.1]], np.float32),
            np.array([[0.1], [0.2]], np.float32),
            np.array([[0.2], [0.3]], np.float32)
        ])

示例#2

文件： policy_gradient_test.py 项目： bigdatasciencegroup/Microsoft-Cognitive-Recognition-Toolkit-CNTK

    def test_rollout_with_update(self, mock_parameters):
        self._setup_parameters(mock_parameters.return_value)
        mock_parameters.return_value.update_frequency = 2

        action_space = spaces.Discrete(2)
        observation_space = spaces.Box(0, 1, (1, ))
        sut = ActorCritic('', observation_space, action_space)
        sut._update_networks = MagicMock()

        sut._choose_action = Mock(
            side_effect=[(0, ''), (1, ''), (1, ''), (0, ''), (1, ''), (0, '')])

        sut.start(np.array([0.1], np.float32))
        sut.step(0.1, np.array([0.2], np.float32))
        self.assertEqual(sut._trajectory_rewards, [0.1])
        self.assertEqual(sut._trajectory_actions, [0, 1])
        self.assertEqual(sut._trajectory_states, [0.1, 0.2])
        self.assertEqual(sut._update_networks.call_count, 0)

        sut.step(0.2, np.array([0.3], np.float32))
        self.assertEqual(sut._trajectory_rewards, [])
        self.assertEqual(sut._trajectory_actions, [1])
        self.assertEqual(sut._trajectory_states, [0.3])
        self.assertEqual(sut._update_networks.call_count, 1)

        sut.step(0.3, np.array([0.4], np.float32))
        self.assertEqual(sut._trajectory_rewards, [0.3])
        self.assertEqual(sut._trajectory_actions, [1, 0])
        self.assertEqual(sut._trajectory_states, [0.3, 0.4])
        self.assertEqual(sut._update_networks.call_count, 1)

        sut.start(np.array([0.5], np.float32))
        self.assertEqual(sut._trajectory_rewards, [])
        self.assertEqual(sut._trajectory_actions, [1])
        self.assertEqual(sut._trajectory_states, [0.5])
        self.assertEqual(sut._update_networks.call_count, 1)

        sut.step(0.4, np.array([0.6], np.float32))
        self.assertEqual(sut._trajectory_rewards, [])
        self.assertEqual(sut._trajectory_actions, [0])
        self.assertEqual(sut._trajectory_states, [0.6])
        self.assertEqual(sut._update_networks.call_count, 2)

        sut.end(0.5, np.array([0.7], np.float32))
        self.assertEqual(sut._trajectory_rewards, [0.5])
        self.assertEqual(sut._trajectory_actions, [0])
        self.assertEqual(sut._trajectory_states, [0.6])
        self.assertEqual(sut._update_networks.call_count, 2)

示例#3

文件： policy_gradient_test.py 项目： AllanYiin/CNTK

    def test_rollout_with_update(self, mock_parameters):
        self._setup_parameters(mock_parameters.return_value)
        mock_parameters.return_value.update_frequency = 2

        action_space = spaces.Discrete(2)
        observation_space = spaces.Box(0, 1, (1,))
        sut = ActorCritic('', observation_space, action_space)
        sut._update_networks = MagicMock()

        sut._choose_action = Mock(side_effect=[
            (0, ''), (1, ''), (1, ''), (0, ''), (1, ''), (0, '')])

        sut.start(np.array([0.1], np.float32))
        sut.step(0.1, np.array([0.2], np.float32))
        self.assertEqual(sut._trajectory_rewards, [0.1])
        self.assertEqual(sut._trajectory_actions, [0, 1])
        self.assertEqual(sut._trajectory_states, [0.1, 0.2])
        self.assertEqual(sut._update_networks.call_count, 0)

        sut.step(0.2, np.array([0.3], np.float32))
        self.assertEqual(sut._trajectory_rewards, [])
        self.assertEqual(sut._trajectory_actions, [1])
        self.assertEqual(sut._trajectory_states, [0.3])
        self.assertEqual(sut._update_networks.call_count, 1)

        sut.step(0.3, np.array([0.4], np.float32))
        self.assertEqual(sut._trajectory_rewards, [0.3])
        self.assertEqual(sut._trajectory_actions, [1, 0])
        self.assertEqual(sut._trajectory_states, [0.3, 0.4])
        self.assertEqual(sut._update_networks.call_count, 1)

        sut.start(np.array([0.5], np.float32))
        self.assertEqual(sut._trajectory_rewards, [])
        self.assertEqual(sut._trajectory_actions, [1])
        self.assertEqual(sut._trajectory_states, [0.5])
        self.assertEqual(sut._update_networks.call_count, 1)

        sut.step(0.4, np.array([0.6], np.float32))
        self.assertEqual(sut._trajectory_rewards, [])
        self.assertEqual(sut._trajectory_actions, [0])
        self.assertEqual(sut._trajectory_states, [0.6])
        self.assertEqual(sut._update_networks.call_count, 2)

        sut.end(0.5, np.array([0.7], np.float32))
        self.assertEqual(sut._trajectory_rewards, [0.5])
        self.assertEqual(sut._trajectory_actions, [0])
        self.assertEqual(sut._trajectory_states, [0.6])
        self.assertEqual(sut._update_networks.call_count, 2)

示例#4

文件： policy_gradient_test.py 项目： bigdatasciencegroup/Microsoft-Cognitive-Recognition-Toolkit-CNTK

    def test_rollout(self):
        action_space = spaces.Discrete(2)
        observation_space = spaces.Box(0, 1, (1, ))
        sut = ActorCritic('', observation_space, action_space)

        sut._choose_action = Mock(side_effect=[(0, ''), (1, ''), (1, '')])

        sut.start(np.array([0.1], np.float32))
        sut.step(0.1, np.array([0.2], np.float32))
        sut.step(0.2, np.array([0.3], np.float32))

        self.assertEqual(sut._trajectory_rewards, [0.1, 0.2])
        self.assertEqual(sut._trajectory_actions, [0, 1, 1])
        self.assertEqual(sut._trajectory_states, [0.1, 0.2, 0.3])

        sut.end(0.3, np.array([0.4], np.float32))

        self.assertEqual(sut._trajectory_rewards, [0.1, 0.2, 0.3])
        self.assertEqual(sut._trajectory_actions, [0, 1, 1])
        self.assertEqual(sut._trajectory_states, [0.1, 0.2, 0.3])

示例#5

文件： policy_gradient_test.py 项目： AllanYiin/CNTK

    def test_rollout(self):
        action_space = spaces.Discrete(2)
        observation_space = spaces.Box(0, 1, (1,))
        sut = ActorCritic('', observation_space, action_space)

        sut._choose_action = Mock(side_effect=[(0, ''), (1, ''), (1, '')])

        sut.start(np.array([0.1], np.float32))
        sut.step(0.1, np.array([0.2], np.float32))
        sut.step(0.2, np.array([0.3], np.float32))

        self.assertEqual(sut._trajectory_rewards, [0.1, 0.2])
        self.assertEqual(sut._trajectory_actions, [0, 1, 1])
        self.assertEqual(sut._trajectory_states, [0.1, 0.2, 0.3])

        sut.end(0.3, np.array([0.4], np.float32))

        self.assertEqual(sut._trajectory_rewards, [0.1, 0.2, 0.3])
        self.assertEqual(sut._trajectory_actions, [0, 1, 1])
        self.assertEqual(sut._trajectory_states, [0.1, 0.2, 0.3])

示例#6

文件： policy_gradient_test.py 项目： AllanYiin/CNTK

    def test_rollout_preprocess(self, mock_parameters):
        self._setup_parameters(mock_parameters.return_value)
        mock_parameters.return_value.preprocessing = \
            'cntk.contrib.deeprl.agent.shared.preprocessing.SlidingWindow'
        mock_parameters.return_value.preprocessing_args = '(2, "float32")'

        action_space = spaces.Discrete(2)
        observation_space = spaces.Box(0, 1, (1,))
        sut = ActorCritic('', observation_space, action_space)

        sut._choose_action = Mock(side_effect=[(0, ''), (1, ''), (1, '')])

        sut.start(np.array([0.1], np.float32))
        sut.step(0.1, np.array([0.2], np.float32))
        sut.step(0.2, np.array([0.3], np.float32))

        self.assertEqual(sut._trajectory_rewards, [0.1, 0.2])
        self.assertEqual(sut._trajectory_actions, [0, 1, 1])
        np.testing.assert_array_equal(
            sut._trajectory_states,
            [
                np.array([[0], [0.1]], np.float32),
                np.array([[0.1], [0.2]], np.float32),
                np.array([[0.2], [0.3]], np.float32)
            ])

        sut.end(0.3, np.array([0.4], np.float32))

        self.assertEqual(sut._trajectory_rewards, [0.1, 0.2, 0.3])
        self.assertEqual(sut._trajectory_actions, [0, 1, 1])
        np.testing.assert_array_equal(
            sut._trajectory_states,
            [
                np.array([[0], [0.1]], np.float32),
                np.array([[0.1], [0.2]], np.float32),
                np.array([[0.2], [0.3]], np.float32)
            ])