add TD3 (#175)

* add TD3

* update

* yapf.....

* Update train.py

examples/TD3/README.md

+## Reproduce TD3 with PARL
+Based on PARL, the TD3 algorithm of deep reinforcement learning has been reproduced, reaching the same level of indicators as the paper in Mujoco benchmarks.
+
+Include following approaches:
++ Clipped Double Q-learning
++ Target Networks and Delayed Policy Update
++ Target Policy Smoothing Regularization
+
+> TD3 in
+[Addressing Function Approximation Error in Actor-Critic Methods](https://arxiv.org/abs/1802.09477)
+
+### Mujoco games introduction
+Please see [here](https://github.com/openai/mujoco-py) to know more about Mujoco games.
+
+### Benchmark result
+
+<img src=".benchmark/merge.png" width = "1500" height ="260" alt="Performance" />
+
+## How to use
+### Dependencies:
++ python3.5+
++ [paddlepaddle>=1.5.1](https://github.com/PaddlePaddle/Paddle)
++ [parl](https://github.com/PaddlePaddle/PARL)
++ gym
++ mujoco-py>=1.50.1.0
+
+### Start Training:
+```
+# To train an agent for HalfCheetah-v2 game
+python train.py
+
+# To train for different game and different loss type
+# python train.py --env [ENV_NAME]

examples/TD3/mujoco_agent.py

+#   Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import numpy as np
+import parl
+from parl import layers
+from paddle import fluid
+
+
+class MujocoAgent(parl.Agent):
+    def __init__(self, algorithm, obs_dim, act_dim):
+        assert isinstance(obs_dim, int)
+        assert isinstance(act_dim, int)
+        self.obs_dim = obs_dim
+        self.act_dim = act_dim
+        super(MujocoAgent, self).__init__(algorithm)
+
+        # Attention: In the beginning, sync target model totally.
+        self.alg.sync_target(decay=0)
+        self.learn_it = 0
+        self.policy_freq = self.alg.policy_freq
+
+    def build_program(self):
+        self.pred_program = fluid.Program()
+        self.actor_learn_program = fluid.Program()
+        self.critic_learn_program = fluid.Program()
+
+        with fluid.program_guard(self.pred_program):
+            obs = layers.data(
+                name='obs', shape=[self.obs_dim], dtype='float32')
+            self.pred_act = self.alg.predict(obs)
+
+        with fluid.program_guard(self.actor_learn_program):
+            obs = layers.data(
+                name='obs', shape=[self.obs_dim], dtype='float32')
+            self.actor_cost = self.alg.actor_learn(obs)
+
+        with fluid.program_guard(self.critic_learn_program):
+            obs = layers.data(
+                name='obs', shape=[self.obs_dim], dtype='float32')
+            act = layers.data(
+                name='act', shape=[self.act_dim], dtype='float32')
+            reward = layers.data(name='reward', shape=[], dtype='float32')
+            next_obs = layers.data(
+                name='next_obs', shape=[self.obs_dim], dtype='float32')
+            terminal = layers.data(name='terminal', shape=[], dtype='bool')
+            self.critic_cost = self.alg.critic_learn(obs, act, reward,
+                                                     next_obs, terminal)
+
+    def predict(self, obs):
+        obs = np.expand_dims(obs, axis=0)
+        act = self.fluid_executor.run(
+            self.pred_program, feed={'obs': obs},
+            fetch_list=[self.pred_act])[0]
+        return act
+
+    def learn(self, obs, act, reward, next_obs, terminal):
+        self.learn_it += 1
+        feed = {
+            'obs': obs,
+            'act': act,
+            'reward': reward,
+            'next_obs': next_obs,
+            'terminal': terminal
+        }
+        critic_cost = self.fluid_executor.run(
+            self.critic_learn_program,
+            feed=feed,
+            fetch_list=[self.critic_cost])[0]
+
+        actor_cost = None
+        if self.learn_it % self.policy_freq == 0:
+            actor_cost = self.fluid_executor.run(
+                self.actor_learn_program,
+                feed={'obs': obs},
+                fetch_list=[self.actor_cost])[0]
+            self.alg.sync_target()
+        return actor_cost, critic_cost

examples/TD3/mujoco_model.py

+#   Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import paddle.fluid as fluid
+import parl
+from parl import layers
+
+
+class MujocoModel(parl.Model):
+    def __init__(self, act_dim, max_action):
+        self.actor_model = ActorModel(act_dim, max_action)
+        self.critic_model = CriticModel()
+
+    def policy(self, obs):
+        return self.actor_model.policy(obs)
+
+    def value(self, obs, act):
+        return self.critic_model.value(obs, act)
+
+    def Q1(self, obs, act):
+        return self.critic_model.Q1(obs, act)
+
+    def get_actor_params(self):
+        return self.actor_model.parameters()
+
+
+class ActorModel(parl.Model):
+    def __init__(self, act_dim, max_action):
+        hid1_size = 400
+        hid2_size = 300
+
+        self.fc1 = layers.fc(size=hid1_size, act='relu')
+        self.fc2 = layers.fc(size=hid2_size, act='relu')
+        self.fc3 = layers.fc(size=act_dim, act='tanh')
+
+        self.max_action = max_action
+
+    def policy(self, obs):
+        hid1 = self.fc1(obs)
+        hid2 = self.fc2(hid1)
+        means = self.fc3(hid2)
+        means = means * self.max_action
+        return means
+
+
+class CriticModel(parl.Model):
+    def __init__(self):
+        hid1_size = 400
+        hid2_size = 300
+
+        self.fc1 = layers.fc(size=hid1_size, act='relu')
+        self.fc2 = layers.fc(size=hid2_size, act='relu')
+        self.fc3 = layers.fc(size=1, act=None)
+
+        self.fc4 = layers.fc(size=hid1_size, act='relu')
+        self.fc5 = layers.fc(size=hid2_size, act='relu')
+        self.fc6 = layers.fc(size=1, act=None)
+
+    def value(self, obs, act):
+        hid1 = self.fc1(obs)
+        concat1 = layers.concat([hid1, act], axis=1)
+        Q1 = self.fc2(concat1)
+        Q1 = self.fc3(Q1)
+        Q1 = layers.squeeze(Q1, axes=[1])
+
+        hid2 = self.fc4(obs)
+        concat2 = layers.concat([hid2, act], axis=1)
+        Q2 = self.fc5(concat2)
+        Q2 = self.fc6(Q2)
+        Q2 = layers.squeeze(Q2, axes=[1])
+        return Q1, Q2
+
+    def Q1(self, obs, act):
+        hid1 = self.fc1(obs)
+        concat1 = layers.concat([hid1, act], axis=1)
+        Q1 = self.fc2(concat1)
+        Q1 = self.fc3(Q1)
+        Q1 = layers.squeeze(Q1, axes=[1])
+
+        return Q1

examples/TD3/train.py

+#   Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import argparse
+import gym
+import numpy as np
+import time
+import parl
+from mujoco_agent import MujocoAgent
+from mujoco_model import MujocoModel
+from parl.utils import logger, tensorboard, action_mapping, ReplayMemory
+
+MAX_EPISODES = 5000
+ACTOR_LR = 3e-4
+CRITIC_LR = 3e-4
+GAMMA = 0.99
+TAU = 0.005
+MEMORY_SIZE = int(1e6)
+WARMUP_SIZE = 1e4
+BATCH_SIZE = 256
+ENV_SEED = 1
+EXPL_NOISE = 0.1  # Std of Gaussian exploration noise
+
+
+def run_train_episode(env, agent, rpm):
+    obs = env.reset()
+    total_reward = 0
+    steps = 0
+    max_action = float(env.action_space.high[0])
+    while True:
+        steps += 1
+        batch_obs = np.expand_dims(obs, axis=0)
+
+        if rpm.size() < WARMUP_SIZE:
+            action = env.action_space.sample()
+        else:
+            action = agent.predict(batch_obs.astype('float32'))
+            action = np.squeeze(action)
+
+            # Add exploration noise, and clip to [-max_action, max_action]
+            action = np.clip(
+                np.random.normal(action, EXPL_NOISE * max_action), -max_action,
+                max_action)
+
+        next_obs, reward, done, info = env.step(action)
+
+        rpm.append(obs, action, reward, next_obs, done)
+
+        if rpm.size() > WARMUP_SIZE:
+            batch_obs, batch_action, batch_reward, batch_next_obs, batch_terminal = rpm.sample_batch(
+                BATCH_SIZE)
+            agent.learn(batch_obs, batch_action, batch_reward, batch_next_obs,
+                        batch_terminal)
+
+        obs = next_obs
+        total_reward += reward
+
+        if done:
+            break
+    return total_reward, steps
+
+
+def run_evaluate_episode(env, agent):
+    obs = env.reset()
+    total_reward = 0
+    while True:
+        batch_obs = np.expand_dims(obs, axis=0)
+        action = agent.predict(batch_obs.astype('float32'))
+        action = np.squeeze(action)
+        action = action_mapping(action, env.action_space.low[0],
+                                env.action_space.high[0])
+
+        next_obs, reward, done, info = env.step(action)
+
+        obs = next_obs
+        total_reward += reward
+
+        if done:
+            break
+    return total_reward
+
+
+def main():
+    env = gym.make(args.env)
+    env.seed(ENV_SEED)
+
+    obs_dim = env.observation_space.shape[0]
+    act_dim = env.action_space.shape[0]
+    max_action = float(env.action_space.high[0])
+
+    model = MujocoModel(act_dim, max_action)
+    algorithm = parl.algorithms.TD3(
+        model,
+        max_action=max_action,
+        gamma=GAMMA,
+        tau=TAU,
+        actor_lr=ACTOR_LR,
+        critic_lr=CRITIC_LR)
+    agent = MujocoAgent(algorithm, obs_dim, act_dim)
+
+    rpm = ReplayMemory(MEMORY_SIZE, obs_dim, act_dim)
+
+    test_flag = 0
+    total_steps = 0
+    while total_steps < args.train_total_steps:
+        train_reward, steps = run_train_episode(env, agent, rpm)
+        total_steps += steps
+        logger.info('Steps: {} Reward: {}'.format(total_steps, train_reward))
+        tensorboard.add_scalar('train/episode_reward', train_reward,
+                               total_steps)
+
+        if total_steps // args.test_every_steps >= test_flag:
+            while total_steps // args.test_every_steps >= test_flag:
+                test_flag += 1
+            evaluate_reward = run_evaluate_episode(env, agent)
+            logger.info('Steps {}, Evaluate reward: {}'.format(
+                total_steps, evaluate_reward))
+            tensorboard.add_scalar('eval/episode_reward', evaluate_reward,
+                                   total_steps)
+
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        '--env', help='Mujoco environment name', default='HalfCheetah-v2')
+    parser.add_argument(
+        '--train_total_steps',
+        type=int,
+        default=int(1e7),
+        help='maximum training steps')
+    parser.add_argument(
+        '--test_every_steps',
+        type=int,
+        default=int(1e4),
+        help='the step interval between two consecutive evaluations')
+
+    args = parser.parse_args()
+
+    main()

parl/algorithms/fluid/__init__.py

@@ -18,4 +18,5 @@ from parl.algorithms.fluid.dqn import *
 from parl.algorithms.fluid.ddqn import *
 from parl.algorithms.fluid.policy_gradient import *
 from parl.algorithms.fluid.ppo import *
+from parl.algorithms.fluid.td3 import *
 from parl.algorithms.fluid.impala.impala import *

parl/algorithms/fluid/td3.py

+#   Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from parl.core.fluid import layers
+from copy import deepcopy
+from paddle import fluid
+from parl.core.fluid.algorithm import Algorithm
+
+__all__ = ['TD3']
+
+
+class TD3(Algorithm):
+    def __init__(
+            self,
+            model,
+            max_action,
+            gamma=None,
+            tau=None,
+            actor_lr=None,
+            critic_lr=None,
+            policy_noise=0.2,  # Noise added to target policy during critic update
+            noise_clip=0.5,  # Range to clip target policy noise
+            policy_freq=2):  # Frequency of delayed policy updates
+        assert isinstance(gamma, float)
+        assert isinstance(tau, float)
+        assert isinstance(actor_lr, float)
+        assert isinstance(critic_lr, float)
+        self.max_action = max_action
+        self.gamma = gamma
+        self.tau = tau
+        self.actor_lr = actor_lr
+        self.critic_lr = critic_lr
+        self.policy_noise = policy_noise
+        self.noise_clip = noise_clip
+        self.policy_freq = policy_freq
+
+        self.model = model
+        self.target_model = deepcopy(model)
+
+    def predict(self, obs):
+        """ use actor model of self.model to predict the action
+        """
+        return self.model.policy(obs)
+
+    def actor_learn(self, obs):
+        action = self.model.policy(obs)
+        Q = self.model.Q1(obs, action)
+        cost = layers.reduce_mean(-1.0 * Q)
+        optimizer = fluid.optimizer.AdamOptimizer(self.actor_lr)
+        optimizer.minimize(cost, parameter_list=self.model.get_actor_params())
+        return cost
+
+    def critic_learn(self, obs, action, reward, next_obs, terminal):
+        noise = layers.gaussian_random_batch_size_like(
+            action, shape=[-1, action.shape[1]])
+        noise = layers.clip(
+            noise * self.policy_noise,
+            min=-self.noise_clip,
+            max=self.noise_clip)
+        next_action = self.target_model.policy(next_obs) + noise
+        next_action = layers.clip(next_action, -self.max_action,
+                                  self.max_action)
+
+        next_Q1, next_Q2 = self.target_model.value(next_obs, next_action)
+        next_Q = layers.elementwise_min(next_Q1, next_Q2)
+
+        terminal = layers.cast(terminal, dtype='float32')
+        target_Q = reward + (1.0 - terminal) * self.gamma * next_Q
+        target_Q.stop_gradient = True
+
+        current_Q1, current_Q2 = self.model.value(obs, action)
+        cost = layers.square_error_cost(current_Q1,
+                                        target_Q) + layers.square_error_cost(
+                                            current_Q2, target_Q)
+        cost = layers.reduce_mean(cost)
+        optimizer = fluid.optimizer.AdamOptimizer(self.critic_lr)
+        optimizer.minimize(cost)
+        return cost
+
+    def sync_target(self, decay=None):
+        if decay is None:
+            decay = 1.0 - self.tau
+        self.model.sync_weights_to(self.target_model, decay=decay)