[Doc] References to TED (#1839)

Author: Vincent Moens

docs/source/reference/data.rst

@@ -301,6 +301,8 @@ common setting when using replay buffers.
 TorchRL Episode Data Format (TED)
 ---------------------------------
 
+.. _TED-format:
+
 In TorchRL, sequential data is consistently presented in a specific format, known
 as the TorchRL Episode Data Format (TED). This format is crucial for the seamless
 integration and functioning of various components within TorchRL.

torchrl/data/datasets/atari_dqn.py

@@ -41,7 +41,7 @@ class AtariDQNExperienceReplay(TensorDictReplayBuffer):
     The sub-sampling rate (frame-skip) is equal to 4, meaning that each game dataset
     has 50 million steps in total.
 
-    The data format follows the TED convention. Since the dataset is quite heavy,
+    The data format follows the :ref:`TED convention <TED-format>`. Since the dataset is quite heavy,
     the data formatting is done on-line, at sampling time.
 
     To make training more modular, we split the dataset in each of the Atari games

torchrl/data/datasets/d4rl.py

@@ -38,6 +38,7 @@ class D4RLExperienceReplay(TensorDictReplayBuffer):
     To install D4RL, follow the instructions on the
     `official repo <https://github.com/Farama-Foundation/D4RL>`__.
 
+    The data format follows the :ref:`TED convention <TED-format>`.
     The replay buffer contains the env specs under D4RLExperienceReplay.specs.
 
     If present, metadata will be written in ``D4RLExperienceReplay.metadata``

torchrl/data/datasets/gen_dgrl.py

@@ -34,6 +34,8 @@ class GenDGRLExperienceReplay(TensorDictReplayBuffer):
 
     GitHub: https://github.com/facebookresearch/gen_dgrl
 
+    The data format follows the :ref:`TED convention <TED-format>`.
+
     This class gives you access to the ProcGen dataset. Each `dataset_id` registered
     in `GenDGRLExperienceReplay.available_datasets` consists in a particular task
     (`"bigfish"`, `"bossfight"`, ...) separated from a category (`"1M_E"`, `"1M_S"`, ...)

torchrl/data/datasets/minari_data.py

@@ -59,6 +59,10 @@
 class MinariExperienceReplay(TensorDictReplayBuffer):
     """Minari Experience replay dataset.
 
+    Learn more about Minari on their website: https://minari.farama.org/
+
+    The data format follows the :ref:`TED convention <TED-format>`.
+
     Args:
         dataset_id (str): The dataset to be downloaded. Must be part of MinariExperienceReplay.available_datasets
         batch_size (int): Batch-size used during sampling. Can be overridden by `data.sample(batch_size)` if

torchrl/data/datasets/openml.py

@@ -27,6 +27,8 @@ class OpenMLExperienceReplay(TensorDictReplayBuffer):
     This class provides an easy entry point for public datasets.
     See "Dua, D. and Graff, C. (2017) UCI Machine Learning Repository. http://archive.ics.uci.edu/ml"
 
+    The data format follows the :ref:`TED convention <TED-format>`.
+
     The data is accessed via scikit-learn. Make sure sklearn and pandas are
     installed before retrieving the data:
 

torchrl/data/datasets/openx.py

@@ -42,6 +42,8 @@ class OpenXExperienceReplay(TensorDictReplayBuffer):
 
     Paper: https://arxiv.org/abs/2310.08864
 
+    The data format follows the :ref:`TED convention <TED-format>`.
+
     .. note::
         Non-tensor data will be written in the tensordict data using the
         :class:`~tensordict.tensorclass.NonTensorData` primitive.

torchrl/data/datasets/roboset.py

@@ -43,6 +43,8 @@ class RobosetExperienceReplay(TensorDictReplayBuffer):
 
     Learn more about roboset here: https://sites.google.com/view/robohive/roboset
 
+    The data format follows the :ref:`TED convention <TED-format>`.
+
     Args:
         dataset_id (str): the dataset to be downloaded. Must be part of RobosetExperienceReplay.available_datasets.
         batch_size (int): Batch-size used during sampling. Can be overridden by `data.sample(batch_size)` if

torchrl/data/datasets/vd4rl.py

@@ -50,6 +50,8 @@ class VD4RLExperienceReplay(TensorDictReplayBuffer):
     that is not reward, done-state, action or pixels is moved under a `"state"`
     node.
 
+    The data format follows the :ref:`TED convention <TED-format>`.
+
     Args:
         dataset_id (str): the dataset to be downloaded. Must be part of
             VD4RLExperienceReplay.available_datasets.