Updated CSVLogger so that it no longer overwrites old csv logs

src/myrtlespeech/run/callbacks/csv_logger.py

@@ -9,16 +9,20 @@
 
 
 class CSVLogger(Callback):
-    r"""Logs at the end of an epoch to the file at ``path`` in CSV format.
+    r"""Logs at the end of an epoch in CSV format.
+
+    This callback saves the CSV to ``log_dir`` with file name
+    ``log_<datetime.isoformat>.csv`` when
+    :py:meth:`.CallbackHandler.on_train_begin` is called.
 
     The first entry in each row of the CSV file denotes the date and time the
     entry was logged in ISO 8601 format (:py:meth:`datetime.isoformat`). The
     second entry in each row of the CSV file will be a string denoting the
     current stage. This will either be ``train`` or ``eval``.
 
     Args:
-        path: Path of a file to write values to. The file will be truncated
-            when :py:meth:`.CallbackHandler.on_train_begin` is called.
+        log_dir: A pathlike object representing the directory to which results
+            will be logged.
 
         keys: Keys in :py:data:`.CallbackHandler.state_dict`, passed as
             ``**kwargs`` to the ``CSVLogger``, whose values will be logged. If
@@ -36,13 +40,14 @@ class CSVLogger(Callback):
     Example:
         >>> # imports
         >>> import tempfile
+        >>> import glob
         >>> from myrtlespeech.run.callbacks.callback import CallbackHandler
         >>>
         >>> # create file to write to
-        >>> temp = tempfile.NamedTemporaryFile()
+        >>> temp = tempfile.TemporaryDirectory()
         >>>
         >>> # initialize CSVLogger and CallbackHandler
-        >>> csv_logger = CSVLogger(path=temp.name, keys=["epoch"])
+        >>> csv_logger = CSVLogger(log_dir=temp.name, keys=["epoch"])
         >>> cb_handler = CallbackHandler(callbacks=[csv_logger])
         >>>
         >>> # simulate training and eval example
@@ -56,7 +61,8 @@ class CSVLogger(Callback):
         >>> # note the first entry in each CSV row is the date
         >>> # this will change per each doctest invocation so it is ignored
         >>> # using ellipsis (...)
-        >>> csv_contents = open(temp.name, 'r').read()
+        >>> csv_name = glob.glob(f'{temp.name}/*.csv')[0]
+        >>> csv_contents = open(csv_name, 'r').read()
         >>> print("example_output:\n", csv_contents)  # doctest:+ELLIPSIS
         example_output:
         ...,stage,epoch
@@ -67,12 +73,14 @@ class CSVLogger(Callback):
 
     def __init__(
         self,
-        path: Union[str, pathlib.Path],
+        log_dir: Union[str, pathlib.Path],
         keys: Optional[Container] = None,
         exclude: Optional[Container] = None,
     ):
         super().__init__()
-        self.path = pathlib.Path(path)
+        self.path = (
+            pathlib.Path(log_dir) / f"log_{datetime.now().isoformat()}.csv"
+        )
         self.keys = keys
         self.exclude = set() if exclude is None else exclude
 

src/myrtlespeech/run/run.py

@@ -229,9 +229,7 @@ def run() -> None:
     if args.stop_epoch_after is not None:
         callbacks.append(StopEpochAfter(epoch_batches=args.stop_epoch_after))
 
-    callbacks.extend(
-        [CSVLogger(log_dir.joinpath("log.csv")), Saver(log_dir, seq_to_seq)]
-    )
+    callbacks.extend([CSVLogger(log_dir), Saver(log_dir, seq_to_seq)])
 
     # train and evaluate
     fit(