thu-ml · Trinkle23897 · May 6, 2021 · Apr 23, 2021 · Apr 25, 2021 · Apr 25, 2021
diff --git a/docs/tutorials/cheatsheet.rst b/docs/tutorials/cheatsheet.rst
@@ -30,6 +30,34 @@ Customize Training Process
 See :ref:`customized_trainer`.
 
 
+.. _resume_training:
+
+Resume Training Process
+-----------------------
+
+This is related to `Issue 349 <https://github.com/thu-ml/tianshou/issues/349>`_.
+
+To resume training process from an existing checkpoint, you need to do the following things in the training process:
+
+1. Make sure you write ``save_checkpoint_fn`` which saves everything needed in the training process, i.e., policy, optim, buffer; pass it to trainer;
+2. Use ``BasicLogger`` which contains a tensorboard;
+3. To adjust the save frequency, specify ``save_interval`` when initializing BasicLogger.
+
+And to successfully resume from a checkpoint:
+
+1. Load everything needed in the training process **before trainer initialization**, i.e., policy, optim, buffer;
+2. Set ``resume_from_log=True`` with trainer;
+
+We provide an example to show how these steps work: checkout `test_c51.py <https://github.com/thu-ml/tianshou/blob/master/test/discrete/test_c51.py>`_, `test_ppo.py <https://github.com/thu-ml/tianshou/blob/master/test/continuous/test_ppo.py>`_ or `test_il_bcq.py <https://github.com/thu-ml/tianshou/blob/master/test/discrete/test_il_bcq.py>`_ by running
+
+.. code-block:: console
+
+    $ python3 test/discrete/test_c51.py  # train some epoch
+    $ python3 test/discrete/test_c51.py --resume  # restore from existing log and continuing training
+
+
+To correctly render the data (including several tfevent files), we highly recommend using ``tensorboard >= 2.5.0`` (see `here <https://github.com/thu-ml/tianshou/pull/350#issuecomment-829123378>`_ for the reason). Otherwise, it may cause overlapping issue that you need to manually handle with.
+
 .. _parallel_sampling:
 
 Parallel Sampling

diff --git a/examples/atari/atari_bcq.py b/examples/atari/atari_bcq.py
@@ -85,8 +85,7 @@ def test_discrete_bcq(args=get_args()):
         feature_net, args.action_shape, device=args.device,
         hidden_sizes=args.hidden_sizes, softmax_output=False).to(args.device)
     optim = torch.optim.Adam(
-        set(policy_net.parameters()).union(imitation_net.parameters()),
-        lr=args.lr)
+        list(policy_net.parameters()) + list(imitation_net.parameters()), lr=args.lr)
     # define policy
     policy = DiscreteBCQPolicy(
         policy_net, imitation_net, optim, args.gamma, args.n_step,

diff --git a/examples/mujoco/mujoco_a2c.py b/examples/mujoco/mujoco_a2c.py
@@ -101,7 +101,7 @@ def test_a2c(args=get_args()):
             torch.nn.init.zeros_(m.bias)
             m.weight.data.copy_(0.01 * m.weight.data)
 
-    optim = torch.optim.RMSprop(set(actor.parameters()).union(critic.parameters()),
+    optim = torch.optim.RMSprop(list(actor.parameters()) + list(critic.parameters()),
                                 lr=args.lr, eps=1e-5, alpha=0.99)
 
     lr_scheduler = None

diff --git a/examples/mujoco/mujoco_ppo.py b/examples/mujoco/mujoco_ppo.py
@@ -106,8 +106,8 @@ def test_ppo(args=get_args()):
             torch.nn.init.zeros_(m.bias)
             m.weight.data.copy_(0.01 * m.weight.data)
 
-    optim = torch.optim.Adam(set(
-        actor.parameters()).union(critic.parameters()), lr=args.lr)
+    optim = torch.optim.Adam(
+        list(actor.parameters()) + list(critic.parameters()), lr=args.lr)
 
     lr_scheduler = None
     if args.lr_decay:

diff --git a/setup.py b/setup.py
@@ -48,7 +48,7 @@ def get_version() -> str:
         "gym>=0.15.4",
         "tqdm",
         "numpy>1.16.0",  # https://github.com/numpy/numpy/issues/12793
-        "tensorboard",
+        "tensorboard>=2.5.0",
         "torch>=1.4.0",
         "numba>=0.51.0",
         "h5py>=2.10.0",  # to match tensorflow's minimal requirements

diff --git a/test/continuous/test_ddpg.py b/test/continuous/test_ddpg.py
@@ -105,6 +105,7 @@ def stop_fn(mean_rewards):
         update_per_step=args.update_per_step, stop_fn=stop_fn,
         save_fn=save_fn, logger=logger)
     assert stop_fn(result['best_reward'])
+
     if __name__ == '__main__':
         pprint.pprint(result)
         # Let's watch its performance!

diff --git a/test/continuous/test_npg.py b/test/continuous/test_npg.py
@@ -80,8 +80,7 @@ def test_npg(args=get_args()):
         if isinstance(m, torch.nn.Linear):
             torch.nn.init.orthogonal_(m.weight)
             torch.nn.init.zeros_(m.bias)
-    optim = torch.optim.Adam(set(
-        actor.parameters()).union(critic.parameters()), lr=args.lr)
+    optim = torch.optim.Adam(critic.parameters(), lr=args.lr)
 
     # replace DiagGuassian with Independent(Normal) which is equivalent
     # pass *logits to be consistent with policy.forward

diff --git a/test/continuous/test_ppo.py b/test/continuous/test_ppo.py
@@ -47,6 +47,8 @@ def get_args():
     parser.add_argument('--value-clip', type=int, default=1)
     parser.add_argument('--norm-adv', type=int, default=1)
     parser.add_argument('--recompute-adv', type=int, default=0)
+    parser.add_argument('--resume', action="store_true")
+    parser.add_argument("--save-interval", type=int, default=4)
     args = parser.parse_known_args()[0]
     return args
 
@@ -83,8 +85,8 @@ def test_ppo(args=get_args()):
         if isinstance(m, torch.nn.Linear):
             torch.nn.init.orthogonal_(m.weight)
             torch.nn.init.zeros_(m.bias)
-    optim = torch.optim.Adam(set(
-        actor.parameters()).union(critic.parameters()), lr=args.lr)
+    optim = torch.optim.Adam(
+        list(actor.parameters()) + list(critic.parameters()), lr=args.lr)
 
     # replace DiagGuassian with Independent(Normal) which is equivalent
     # pass *logits to be consistent with policy.forward
@@ -114,21 +116,42 @@ def dist(*logits):
     # log
     log_path = os.path.join(args.logdir, args.task, 'ppo')
     writer = SummaryWriter(log_path)
-    logger = BasicLogger(writer)
+    logger = BasicLogger(writer, save_interval=args.save_interval)
 
     def save_fn(policy):
         torch.save(policy.state_dict(), os.path.join(log_path, 'policy.pth'))
 
     def stop_fn(mean_rewards):
         return mean_rewards >= env.spec.reward_threshold
 
+    def save_checkpoint_fn(epoch, env_step, gradient_step):
+        # see also: https://pytorch.org/tutorials/beginner/saving_loading_models.html
+        torch.save({
+            'model': policy.state_dict(),
+            'optim': optim.state_dict(),
+        }, os.path.join(log_path, 'checkpoint.pth'))
+
+    if args.resume:
+        # load from existing checkpoint
+        print(f"Loading agent under {log_path}")
+        ckpt_path = os.path.join(log_path, 'checkpoint.pth')
+        if os.path.exists(ckpt_path):
+            checkpoint = torch.load(ckpt_path, map_location=args.device)
+            policy.load_state_dict(checkpoint['model'])
+            optim.load_state_dict(checkpoint['optim'])
+            print("Successfully restore policy and optim.")
+        else:
+            print("Fail to restore policy and optim.")
+
     # trainer
     result = onpolicy_trainer(
-        policy, train_collector, test_collector, args.epoch,
-        args.step_per_epoch, args.repeat_per_collect, args.test_num, args.batch_size,
+        policy, train_collector, test_collector, args.epoch, args.step_per_epoch,
+        args.repeat_per_collect, args.test_num, args.batch_size,
         episode_per_collect=args.episode_per_collect, stop_fn=stop_fn, save_fn=save_fn,
-        logger=logger)
+        logger=logger, resume_from_log=args.resume,
+        save_checkpoint_fn=save_checkpoint_fn)
     assert stop_fn(result['best_reward'])
+
     if __name__ == '__main__':
         pprint.pprint(result)
         # Let's watch its performance!
@@ -140,5 +163,10 @@ def stop_fn(mean_rewards):
         print(f"Final reward: {rews.mean()}, length: {lens.mean()}")
 
 
+def test_ppo_resume(args=get_args()):
+    args.resume = True
+    test_ppo(args)
+
+
 if __name__ == '__main__':
     test_ppo()
diff --git a/test/continuous/test_sac_with_il.py b/test/continuous/test_sac_with_il.py
@@ -124,6 +124,7 @@ def stop_fn(mean_rewards):
         update_per_step=args.update_per_step, stop_fn=stop_fn,
         save_fn=save_fn, logger=logger)
     assert stop_fn(result['best_reward'])
+
     if __name__ == '__main__':
         pprint.pprint(result)
         # Let's watch its performance!

diff --git a/test/continuous/test_td3.py b/test/continuous/test_td3.py
@@ -119,6 +119,7 @@ def stop_fn(mean_rewards):
         update_per_step=args.update_per_step, stop_fn=stop_fn,
         save_fn=save_fn, logger=logger)
     assert stop_fn(result['best_reward'])
+
     if __name__ == '__main__':
         pprint.pprint(result)
         # Let's watch its performance!

diff --git a/test/continuous/test_trpo.py b/test/continuous/test_trpo.py
@@ -27,7 +27,8 @@ def get_args():
     parser.add_argument('--epoch', type=int, default=5)
     parser.add_argument('--step-per-epoch', type=int, default=50000)
     parser.add_argument('--step-per-collect', type=int, default=2048)
-    parser.add_argument('--repeat-per-collect', type=int, default=1)
+    parser.add_argument('--repeat-per-collect', type=int,
+                        default=2)  # theoretically it should be 1
     parser.add_argument('--batch-size', type=int, default=99999)
     parser.add_argument('--hidden-sizes', type=int, nargs='*', default=[64, 64])
     parser.add_argument('--training-num', type=int, default=16)
@@ -82,8 +83,7 @@ def test_trpo(args=get_args()):
         if isinstance(m, torch.nn.Linear):
             torch.nn.init.orthogonal_(m.weight)
             torch.nn.init.zeros_(m.bias)
-    optim = torch.optim.Adam(set(
-        actor.parameters()).union(critic.parameters()), lr=args.lr)
+    optim = torch.optim.Adam(critic.parameters(), lr=args.lr)
 
     # replace DiagGuassian with Independent(Normal) which is equivalent
     # pass *logits to be consistent with policy.forward

diff --git a/test/discrete/test_a2c_with_il.py b/test/discrete/test_a2c_with_il.py
@@ -74,8 +74,8 @@ def test_a2c_with_il(args=get_args()):
               device=args.device)
     actor = Actor(net, args.action_shape, device=args.device).to(args.device)
     critic = Critic(net, device=args.device).to(args.device)
-    optim = torch.optim.Adam(set(
-        actor.parameters()).union(critic.parameters()), lr=args.lr)
+    optim = torch.optim.Adam(
+        list(actor.parameters()) + list(critic.parameters()), lr=args.lr)
     dist = torch.distributions.Categorical
     policy = A2CPolicy(
         actor, critic, optim, dist,
@@ -106,6 +106,7 @@ def stop_fn(mean_rewards):
         episode_per_collect=args.episode_per_collect, stop_fn=stop_fn, save_fn=save_fn,
         logger=logger)
     assert stop_fn(result['best_reward'])
+
     if __name__ == '__main__':
         pprint.pprint(result)
         # Let's watch its performance!
@@ -135,6 +136,7 @@ def stop_fn(mean_rewards):
         args.il_step_per_epoch, args.step_per_collect, args.test_num,
         args.batch_size, stop_fn=stop_fn, save_fn=save_fn, logger=logger)
     assert stop_fn(result['best_reward'])
+
     if __name__ == '__main__':
         pprint.pprint(result)
         # Let's watch its performance!

diff --git a/test/discrete/test_c51.py b/test/discrete/test_c51.py
@@ -1,6 +1,7 @@
 import os
 import gym
 import torch
+import pickle
 import pprint
 import argparse
 import numpy as np
@@ -43,9 +44,11 @@ def get_args():
                         action="store_true", default=False)
     parser.add_argument('--alpha', type=float, default=0.6)
     parser.add_argument('--beta', type=float, default=0.4)
+    parser.add_argument('--resume', action="store_true")
     parser.add_argument(
         '--device', type=str,
         default='cuda' if torch.cuda.is_available() else 'cpu')
+    parser.add_argument("--save-interval", type=int, default=4)
     args = parser.parse_known_args()[0]
     return args
 
@@ -90,7 +93,7 @@ def test_c51(args=get_args()):
     # log
     log_path = os.path.join(args.logdir, args.task, 'c51')
     writer = SummaryWriter(log_path)
-    logger = BasicLogger(writer)
+    logger = BasicLogger(writer, save_interval=args.save_interval)
 
     def save_fn(policy):
         torch.save(policy.state_dict(), os.path.join(log_path, 'policy.pth'))
@@ -112,14 +115,42 @@ def train_fn(epoch, env_step):
     def test_fn(epoch, env_step):
         policy.set_eps(args.eps_test)
 
+    def save_checkpoint_fn(epoch, env_step, gradient_step):
+        # see also: https://pytorch.org/tutorials/beginner/saving_loading_models.html
+        torch.save({
+            'model': policy.state_dict(),
+            'optim': optim.state_dict(),
+        }, os.path.join(log_path, 'checkpoint.pth'))
+        pickle.dump(train_collector.buffer,
+                    open(os.path.join(log_path, 'train_buffer.pkl'), "wb"))
+
+    if args.resume:
+        # load from existing checkpoint
+        print(f"Loading agent under {log_path}")
+        ckpt_path = os.path.join(log_path, 'checkpoint.pth')
+        if os.path.exists(ckpt_path):
+            checkpoint = torch.load(ckpt_path, map_location=args.device)
+            policy.load_state_dict(checkpoint['model'])
+            policy.optim.load_state_dict(checkpoint['optim'])
+            print("Successfully restore policy and optim.")
+        else:
+            print("Fail to restore policy and optim.")
+        buffer_path = os.path.join(log_path, 'train_buffer.pkl')
+        if os.path.exists(buffer_path):
+            train_collector.buffer = pickle.load(open(buffer_path, "rb"))
+            print("Successfully restore buffer.")
+        else:
+            print("Fail to restore buffer.")
+
     # trainer
     result = offpolicy_trainer(
         policy, train_collector, test_collector, args.epoch,
         args.step_per_epoch, args.step_per_collect, args.test_num,
         args.batch_size, update_per_step=args.update_per_step, train_fn=train_fn,
-        test_fn=test_fn, stop_fn=stop_fn, save_fn=save_fn, logger=logger)
-
+        test_fn=test_fn, stop_fn=stop_fn, save_fn=save_fn, logger=logger,
+        resume_from_log=args.resume, save_checkpoint_fn=save_checkpoint_fn)
     assert stop_fn(result['best_reward'])
+
     if __name__ == '__main__':
         pprint.pprint(result)
         # Let's watch its performance!
@@ -132,6 +163,11 @@ def test_fn(epoch, env_step):
         print(f"Final reward: {rews.mean()}, length: {lens.mean()}")
 
 
+def test_c51_resume(args=get_args()):
+    args.resume = True
+    test_c51(args)
+
+
 def test_pc51(args=get_args()):
     args.prioritized_replay = True
     args.gamma = .95

diff --git a/test/discrete/test_dqn.py b/test/discrete/test_dqn.py
@@ -120,7 +120,6 @@ def test_fn(epoch, env_step):
         args.step_per_epoch, args.step_per_collect, args.test_num,
         args.batch_size, update_per_step=args.update_per_step, train_fn=train_fn,
         test_fn=test_fn, stop_fn=stop_fn, save_fn=save_fn, logger=logger)
-
     assert stop_fn(result['best_reward'])
 
     if __name__ == '__main__':

diff --git a/test/discrete/test_drqn.py b/test/discrete/test_drqn.py
@@ -99,8 +99,8 @@ def test_fn(epoch, env_step):
         args.batch_size, update_per_step=args.update_per_step,
         train_fn=train_fn, test_fn=test_fn, stop_fn=stop_fn,
         save_fn=save_fn, logger=logger)
-
     assert stop_fn(result['best_reward'])
+
     if __name__ == '__main__':
         pprint.pprint(result)
         # Let's watch its performance!