fix: warning hidden vs rnn size, priority hidden_size

src/rxn/onmt_models/scripts/rxn_onmt_train.py

@@ -1,4 +1,5 @@
 import logging
+import warnings
 from pathlib import Path
 from typing import Tuple
 
@@ -22,6 +23,24 @@ def get_src_tgt_vocab(data: Path) -> Tuple[Path, Path]:
     return src_vocab, tgt_vocab
 
 
+def check_rnn_vs_hidden_size(hidden_size: int, rnn_size: int) -> int:
+    """
+    Helper function that checks wether hidden_size and rnn_size are given, decides which one to use and raises warnings.
+    rnn_size always has a default defaults.RNN_SIZE, if no hidden_size is given, rnn_size will be used.
+    If hidden_size is given, hidden size will be used.
+    """
+    if hidden_size is None:
+        warnings.warn(
+            f"Argument hidden_size is not given, rnn_size with value {rnn_size} will be used"
+        )
+        return rnn_size
+    if hidden_size is not None:
+        warnings.warn(
+            f"Argument hidden_size was given with value {hidden_size}, rnn_size argument will be overwritten."
+        )
+        return hidden_size
+
+
 @click.command(context_settings=dict(show_default=True))
 @click.option("--batch_size", default=defaults.BATCH_SIZE)
 @click.option(
@@ -51,6 +70,7 @@ def get_src_tgt_vocab(data: Path) -> Tuple[Path, Path]:
     help="Directory with OpenNMT-preprocessed files",
 )
 @click.option("--rnn_size", default=defaults.RNN_SIZE)
+@click.option("--hidden_size")
 @click.option("--seed", default=defaults.SEED)
 @click.option("--train_num_steps", default=100000)
 @click.option("--transformer_ff", default=defaults.TRANSFORMER_FF)
@@ -68,6 +88,7 @@ def main(
     no_gpu: bool,
     preprocess_dir: str,
     rnn_size: int,
+    hidden_size: int,
     seed: int,
     train_num_steps: int,
     transformer_ff: int,
@@ -80,6 +101,9 @@ def main(
     `data_weights` parameters are given (Note: needs to be consistent with the
     rxn-onmt-preprocess command executed before training.
     """
+    # Check rnn_size or hidden_size given, not both
+    # NOTE: rnn_size argument is kept for compatibility
+    hidden_size = check_rnn_vs_hidden_size(hidden_size=hidden_size, rnn_size=rnn_size)
 
     # set up paths
     model_files = ModelFiles(model_output_dir)
@@ -110,7 +134,7 @@ def main(
         keep_checkpoint=keep_checkpoint,
         layers=layers,
         learning_rate=learning_rate,
-        hidden_size=rnn_size,
+        hidden_size=hidden_size,
         save_model=model_files.model_prefix,
         seed=seed,
         train_steps=train_num_steps,