workaround for train_set batching during inference time

ufal · Feb 21, 2019 · 0f5649c · 0f5649c
1 parent b384686
commit 0f5649c
Show file tree

Hide file tree

Showing 2 changed files with 5 additions and 1 deletion.
diff --git a/neuralmonkey/learning_utils.py b/neuralmonkey/learning_utils.py
@@ -13,7 +13,7 @@
 from termcolor import colored
 
 from neuralmonkey.logging import log, log_print, warn
-from neuralmonkey.dataset import Dataset
+from neuralmonkey.dataset import Dataset, BatchingScheme
 from neuralmonkey.tf_manager import TensorFlowManager
 from neuralmonkey.runners.base_runner import (
     BaseRunner, ExecutionResult, GraphExecutor, OutputSeries)
@@ -85,6 +85,9 @@ def training_loop(cfg: Namespace) -> None:
                     trainer_result = cfg.tf_manager.execute(
                         batch, feedables, cfg.trainers, train=True,
                         summaries=True)
+                    # workaround: we need to use validation batching scheme
+                    #             during evaluation
+                    batch.batching = BatchingScheme(batch_size=cfg.batch_size)
                     train_results, train_outputs, f_batch = run_on_dataset(
                         cfg.tf_manager, cfg.runners, cfg.dataset_runner, batch,
                         cfg.postprocess, write_out=False)

diff --git a/tests/hier-multiattention.ini b/tests/hier-multiattention.ini
@@ -10,6 +10,7 @@ trainer=<trainer>
 runners=[<runner_hier_noshare_nosentinel>, <runner_hier_share_nosentinel>, <runner_hier_share_sentinel>, <runner_hier_noshare_sentinel>]
 postprocess=None
 evaluation=[("target_hier_noshare_nosentinel", "target", evaluators.BLEU)]
+batch_size=1
 logging_period=1
 validation_period=5
 test_datasets=[<val_data>]