thu-ml · Trinkle23897 · Oct 13, 2021 · Oct 7, 2021 · Oct 13, 2021 · Oct 13, 2021
diff --git a/README.md b/README.md
@@ -194,6 +194,7 @@ buffer_size = 20000
 eps_train, eps_test = 0.1, 0.05
 step_per_epoch, step_per_collect = 10000, 10
 logger = ts.utils.TensorboardLogger(SummaryWriter('log/dqn'))  # TensorBoard is supported!
+# For other loggers: https://tianshou.readthedocs.io/en/master/tutorials/logger.html
 ```
 
 Make environments:

diff --git a/docs/index.rst b/docs/index.rst
@@ -93,6 +93,7 @@ Tianshou is still under development, you can also check out the documents in sta
    tutorials/concepts
    tutorials/batch
    tutorials/tictactoe
+   tutorials/logger
    tutorials/benchmark
    tutorials/cheatsheet
 

diff --git a/docs/spelling_wordlist.txt b/docs/spelling_wordlist.txt
@@ -14,6 +14,7 @@ timestep
 numpy
 ndarray
 stackoverflow
+tensorboard
 len
 tac
 fqf

diff --git a/docs/tutorials/logger.rst b/docs/tutorials/logger.rst
@@ -0,0 +1,62 @@
+Logging Experiments
+===================
+
+Tianshou comes with multiple experiment tracking and logging solutions to manage and reproduce your experiments.
+The dashboard loggers currently available are:
+
+* :class:`~tianshou.utils.TensorboardLogger`
+* :class:`~tianshou.utils.WandbLogger`
+* :class:`~tianshou.utils.LazyLogger`
+
+
+TensorboardLogger
+-----------------
+
+Tensorboard tracks your experiment metrics in a local dashboard. Here is how you can use TensorboardLogger in your experiment:
+
+::
+
+    from torch.utils.tensorboard import SummaryWriter
+    from tianshou.utils import TensorboardLogger
+
+    log_path = os.path.join(args.logdir, args.task, "dqn")
+    writer = SummaryWriter(log_path)
+    writer.add_text("args", str(args))
+    logger = TensorboardLogger(writer)
+    result = trainer(..., logger=logger)
+
+
+WandbLogger
+-----------
+
+:class:`~tianshou.utils.WandbLogger` can be used to visualize your experiments in a hosted `W&B dashboard <https://wandb.ai/home>`_. It can be installed via ``pip install wandb``. You can also save your checkpoints in the cloud and restore your runs from those checkpoints. Here is how you can enable WandbLogger:
+
+::
+
+    from tianshou.utils import WandbLogger
+
+    logger = WandbLogger(...)
+    result = trainer(..., logger=logger)
+
+Please refer to :class:`~tianshou.utils.WandbLogger` documentation for advanced configuration.
+
+For logging checkpoints on any device, you need to define a ``save_checkpoint_fn`` which saves the experiment checkpoint and returns the path of the saved checkpoint:
+
+::
+
+    def save_checkpoint_fn(epoch, env_step, gradient_step):
+        ckpt_path = ...
+        # save model
+        return ckpt_path
+
+Then, use this function with ``WandbLogger`` to automatically version your experiment checkpoints after every ``save_interval`` step.
+
+For resuming runs from checkpoint artifacts on any device, pass the W&B ``run_id`` of the run that you want to continue in ``WandbLogger``. It will then download the latest version of the checkpoint and resume your runs from the checkpoint.
+
+The example scripts are under `test_psrl.py <https://github.com/thu-ml/tianshou/blob/master/test/modelbased/test_psrl.py>`_ and `atari_dqn.py <https://github.com/thu-ml/tianshou/blob/master/examples/atari/atari_dqn.py>`_.
+
+
+LazyLogger
+----------
+
+This is a place-holder logger that does nothing.
-Original file line number
+Diff line change
@@ Expand Up / @@ -14,6 +14,7 @@ timestep @@
     numpy
     ndarray
     stackoverflow
+    tensorboard
     len
     tac
     fqf
@@ Expand Down @@