Add simple predictor creation in code from pretrained model.

combo/models/base.py

@@ -83,7 +83,7 @@ class FeedForwardPredictor(Predictor):
         pred = pred.reshape(-1, CLASSES)
         true = true.reshape(-1)
         mask = mask.reshape(-1)
-        loss = utils.masked_cross_entropy(pred, true, mask) * mask
+        loss = utils.masked_cross_entropy(pred, true, mask)
         loss = loss.reshape(BATCH_SIZE, -1) * sample_weights.unsqueeze(-1)
         return loss.sum() / valid_positions
 

combo/models/lemma.py

@@ -74,12 +74,13 @@ class LemmatizerModel(base.Predictor):
         valid_positions = mask.sum()
         mask = mask.reshape(-1)
         true = true.reshape(-1)
-        loss = utils.masked_cross_entropy(pred, true, mask) * mask
+        loss = utils.masked_cross_entropy(pred, true, mask)
         loss = loss.reshape(BATCH_SIZE, -1) * sample_weights.unsqueeze(-1)
         return loss.sum() / valid_positions
 
     @classmethod
     def from_vocab(cls,
+
                    vocab: data.Vocabulary,
                    char_vocab_namespace: str,
                    lemma_vocab_namespace: str,

combo/models/morpho.py

@@ -63,11 +63,11 @@ class MorphologicalFeatures(base.Predictor):
                 if loss is None:
                     loss = loss_func(pred[:, cat_indices],
                                      true[:, cat_indices].argmax(dim=1),
-                                     mask) * mask
+                                     mask)
                 else:
                     loss += loss_func(pred[:, cat_indices],
                                       true[:, cat_indices].argmax(dim=1),
-                                      mask) * mask
+                                      mask)
         loss = loss.reshape(BATCH_SIZE, -1) * sample_weights.unsqueeze(-1)
         return loss.sum() / valid_positions
 

combo/models/parser.py

@@ -93,7 +93,7 @@ class HeadPredictionModel(base.Predictor):
             pred_i = pred[:, i + 1, :].reshape(BATCH_SIZE, SENTENCE_LENGTH)
             true_i = true[:, i].reshape(-1)
             mask_i = mask[:, i]
-            cross_entropy_loss = utils.masked_cross_entropy(pred_i, true_i, mask_i) * mask_i
+            cross_entropy_loss = utils.masked_cross_entropy(pred_i, true_i, mask_i)
             result.append(cross_entropy_loss)
         cycle_loss = self._cycle_loss(pred)
         loss = torch.stack(result).transpose(1, 0) * sample_weights.unsqueeze(-1)
@@ -162,7 +162,7 @@ class DependencyRelationModel(base.Predictor):
         pred = pred.reshape(-1, DEPENDENCY_RELATIONS)
         true = true.reshape(-1)
         mask = mask.reshape(-1)
-        loss = utils.masked_cross_entropy(pred, true, mask) * mask
+        loss = utils.masked_cross_entropy(pred, true, mask)
         loss = loss.reshape(BATCH_SIZE, -1) * sample_weights.unsqueeze(-1)
         return loss.sum() / valid_positions
 

combo/models/utils.py

@@ -5,4 +5,4 @@ import torch.nn.functional as F
 def masked_cross_entropy(pred: torch.Tensor, true: torch.Tensor, mask: torch.BoolTensor) -> torch.Tensor:
     mask = mask.float().unsqueeze(-1)
     pred = pred + (mask + 1e-45).log()
-    return F.cross_entropy(pred, true, reduction='none')
+    return F.cross_entropy(pred, true, reduction='none') * mask

combo/predict.py

@@ -57,6 +57,9 @@ class SemanticMultitaskPredictor(predictor.Predictor):
         logger.info('Took {} ms'.format((end_time - start_time) * 1000.0))
         return result
 
+    def predict_string(self, sentence: str):
+        return self.predict_json({'sentence': sentence})
+
     @overrides
     def predict_json(self, inputs: common.JsonDict) -> common.JsonDict:
         start_time = time.time()
@@ -139,3 +142,13 @@ class SemanticMultitaskPredictor(predictor.Predictor):
     def with_spacy_tokenizer(cls, model: models.Model,
                              dataset_reader: allen_data.DatasetReader):
         return cls(model, dataset_reader, tokenizers.SpacyTokenizer())
+
+    @classmethod
+    def from_pretrained(cls, path: str, tokenizer=tokenizers.SpacyTokenizer()):
+        util.import_module_and_submodules('combo.commands')
+        util.import_module_and_submodules('combo.models')
+        util.import_module_and_submodules('combo.training')
+        model = models.Model.from_archive(path)
+        dataset_reader = allen_data.DatasetReader.from_params(
+            models.load_archive(path).config['dataset_reader'])
+        return cls(model, dataset_reader, tokenizer)