Fixing temporary bug with inputs and bucketing

when series are bucketed (i.e. batches() do not return data in the same order
as get_series()), inputs were returned in the wrong order.

neuralmonkey/learning_utils.py

@@ -317,6 +317,8 @@ def run_on_dataset(tf_manager: TensorFlowManager,
     feedables = set.union(*[runner.feedables for runner in runners])
     feedables |= dataset_runner.feedables
 
+    fetched_input = {s: [] for s in dataset.series}  # type: Dict[str, List]
+
     processed_examples = 0
     for batch in dataset.batches():
         if 0 < log_progress < time.process_time() - last_log_time:
@@ -335,6 +337,9 @@ def run_on_dataset(tf_manager: TensorFlowManager,
         for script_list, ex_result in zip(batch_results, execution_results):
             script_list.append(ex_result)
 
+        for s_id in batch.series:
+            fetched_input[s_id].extend(batch.get_series(s_id))
+
     # Transpose runner interim results.
     all_results = [join_execution_results(res) for res in batch_results[:-1]]
 
@@ -343,7 +348,6 @@ def run_on_dataset(tf_manager: TensorFlowManager,
     # fetched_input = {
     #     k: [dic[k] for dic in input_transposed] for k in input_transposed[0]}
 
-    fetched_input = {s: list(dataset.get_series(s)) for s in dataset.series}
     fetched_input_lengths = {s: len(fetched_input[s]) for s in dataset.series}
 
     if len(set(fetched_input_lengths.values())) != 1: