bugfix: added cast; plus fixed stupid mistakes in test cases (#426)

* bugfix: added cast; plus fixed stupid mistakes in test cases that let it sneak in

CHANGELOG.md

@@ -10,11 +10,16 @@ Note that Sockeye has checks in place to not translate with an old model that wa
 
 Each version section may have have subsections for: _Added_, _Changed_, _Removed_, _Deprecated_, and _Fixed_.
 
+## [1.18.21]
+### Fixed
+- Constrained decoding was missed a crucial cast
+- Fixed test cases that should have caught this
+
 ## [1.18.20]
 ### Changed
 - Transformer parametrization flags (model size, # of attention heads, feed-forward layer size) can now optionally
   defined separately for encoder & decoder. For example, to use a different transformer model size for the encoder,
-  pass `--transformer-model-size 1024:512`. 
+  pass `--transformer-model-size 1024:512`.
 
 ## [1.18.19]
 ### Added

sockeye/__init__.py

@@ -11,4 +11,4 @@
 # express or implied. See the License for the specific language governing
 # permissions and limitations under the License.
 
-__version__ = '1.18.20'
+__version__ = '1.18.21'

sockeye/inference.py

@@ -1583,7 +1583,7 @@ def _get_best_from_beam(self,
             unmet = mx.nd.array([c.num_needed() if c is not None else 0 for c in constraints], ctx=self.context)
             filtered = mx.nd.where(unmet == 0, seq_scores, self.inf_array)
             filtered = filtered.reshape((self.batch_size, self.beam_size))
-            best_ids += mx.nd.argmin(filtered, axis=1)
+            best_ids += mx.nd.cast(mx.nd.argmin(filtered, axis=1), dtype='int32')
 
         histories = beam_histories if beam_histories is not None else [None] * self.batch_size  # type: List
         return [self._assemble_translation(*x) for x in zip(sequences[best_ids],

test/common.py

@@ -204,7 +204,7 @@ def tmp_digits_dataset(prefix: str,
                     end_pos = min(target_len, 3)
                     constraint = ' '.join(target_words[start_pos:end_pos])
                     new_source['constraints'] = [constraint]
-                    new_sources.append(json.dumps(new_source))
+                new_sources.append(json.dumps(new_source))
 
             with open(data['test_source'], 'w') as out:
                 for json_line in new_sources:

test/integration/test_seq_copy_int.py

@@ -170,7 +170,7 @@ def test_seq_copy(train_params: str,
                             train_target_path=data['target'],
                             dev_source_path=data['validation_source'],
                             dev_target_path=data['validation_target'],
-                            test_source_path=data['test_target'],
+                            test_source_path=data['test_source'],
                             test_target_path=data['test_target'],
                             train_source_factor_paths=train_source_factor_paths,
                             dev_source_factor_paths=dev_source_factor_paths,