Fix up documentation. Add a docstring to TtyrecDataset.

nle/dataset/dataset.py

@@ -36,12 +36,12 @@ def convert_frames(
     :param timestamps: Array of timestamps -  np.array(np.int64) [ SEQ ]
     :param actions: Array of actions at t in response to output at t
         - np.array(np.uint8) [ SEQ ]
+    :param scores: Array of in-game scores -  np.array(np.int32) [ SEQ ]
     :param resets: Array of resets -  np.array(np.uint8) [ SEQ ]
     :param gameids: Array of the gameid of each frame - np.array(np.int32) [ SEQ ]
     :param load_fn: A callback that loads the next file into a converter:
         sig: load_fn(converter) -> bool is_success
 
-    Note actions will only be non-null if the ttyrec read_actions flag is set.
     """
 
     resets[0] = 0
@@ -161,6 +161,46 @@ def __init__(
         subselect_sql=None,
         subselect_sql_args=None,
     ):
+        """
+        An iterable dataset to load minibatches of NetHack games from compressed
+        ttyrec*.bz2 files into numpy arrays. (shape: [batch_size, seq_length, ...])
+
+        This class makes use of a sqlite3 database at `dbfilename` to find the
+        metadata and the location of files in a dataset. It then uses these to
+        create generators which convert the ttyrecs on the fly. Note that the
+        dataset generators always reuse their numpy arrays, writing into the
+        arrays instead of generating new ones. Methods to create and populate the
+        db from an NLE directry can be found in `populate_db.py`.
+
+        Example
+        -------
+            ```
+            import nle.dataset as nld
+
+            if not os.path.exists(nld.db.DB):
+                nld.db.create()
+                nld.populate_db.add_nledata_directory('path/to/nle_data', "data1")
+
+            dataset = nld.TtyrecDataset("data1"):
+
+            for mb in dataset:
+                # NB: dataset reuses np arrays, for performance reasons
+                print(mb)
+            ```
+
+        :param batch_size: Number of parallel games to load.
+        :param seq_length: Number of frames to load per game.
+        :param rows: Row size of the terminal screen.
+        :param cols: Column size of the terminal screen.
+        :param dbfilename: Path to the database file
+        :param gameids: Use a subselection of games (gameids) only.
+        :param shuffle: Shuffle the order of gameids before iterating through them.
+        :param loop_forever: If true, cycle through gameids forever,
+            insted of padding empty batch dims with 0's.
+        :param subselect_sql: SQL Query to subselect games (gameids) using metadata
+        :param subselect_sql_args: SQL Query Args to subselect games (gameids)
+            using metadata.
+        """
         self.batch_size = batch_size
         self.seq_length = seq_length
         self.rows = rows

nle/scripts/read_tty.py

@@ -183,7 +183,6 @@ def main():
                 arrow = "->"
             elif channel == 2:
                 score, *_ = struct.unpack("<i", data)
-                # data = chr(score).encode("ascii", "backslashreplace")
                 data = f"  {score} "
                 arrow = "->"