Add Hyperopt example for BERT classifier

docs/code/data.rst

@@ -130,6 +130,16 @@ Data Loaders
 Data Iterators
 ===============
 
+:hidden:`Batch`
+~~~~~~~~~~~~~~~~~~~~~~~~
+.. autoclass:: texar.torch.data.Batch
+    :members:
+
+:hidden:`FieldBatch`
+~~~~~~~~~~~~~~~~~~~~~~~~
+.. autoclass:: texar.torch.data.FieldBatch
+    :members:
+
 :hidden:`DataIterator`
 ~~~~~~~~~~~~~~~~~~~~~~~~
 .. autoclass:: texar.torch.data.DataIterator

examples/bert/bert_classifier_main.py

@@ -56,7 +56,7 @@
 config_downstream = importlib.import_module(args.config_downstream)
 config_downstream = {
     k: v for k, v in config_downstream.__dict__.items()
-    if not k.startswith('__')}
+    if not k.startswith('__') and k != "hyperparams"}
 
 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
 

examples/bert/bert_classifier_using_executor_main.py

@@ -65,7 +65,7 @@
 config_downstream = importlib.import_module(args.config_downstream)
 config_downstream = {
     k: v for k, v in config_downstream.__dict__.items()
-    if not k.startswith('__')}
+    if not k.startswith('__') and k != "hyperparams"}
 
 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
 

examples/bert/bert_with_hypertuning_main.py

@@ -48,14 +48,6 @@
 parser.add_argument(
     "--checkpoint", type=str, default=None,
     help="Path to a model checkpoint (including bert modules) to restore from.")
-parser.add_argument(
-    "--do-train", action="store_true", help="Whether to run training.")
-parser.add_argument(
-    "--do-eval", action="store_true",
-    help="Whether to run eval on the dev set.")
-parser.add_argument(
-    "--do-test", action="store_true",
-    help="Whether to run test on the test set.")
 args = parser.parse_args()
 
 config_data = importlib.import_module(args.config_data)
@@ -104,7 +96,9 @@ def forward(self,  # type: ignore
         for more details.
 
         Args:
-            `batch`: :class:`texar.data.Batch`
+            `batch`: :class:`texar.data.Batch`. (See
+            https://texar-pytorch.readthedocs.io/en/latest/code/data.html#texar.torch.data.Batch
+            for more details)
                 A batch of inputs to be passed through the model
 
         Returns:
@@ -287,6 +281,9 @@ def objective_func(self, hyperparams: Dict):
             print_model_arch=False
         )
 
+        if args.checkpoint is not None:
+            executor.load(args.checkpoint)
+
         executor.train()
 
         print(f"Loss on the valid dataset "

texar/torch/data/data/dataset_utils.py

@@ -57,9 +57,33 @@ def connect_name(lhs_name, rhs_name):
 
 
 class Batch:
-    r"""Wrapper over Python dictionaries representing a batch. This provides a
-    common interface with :class:`~texar.torch.data.data.dataset_utils.Batch`
-    that allows accessing via attributes.
+    r"""Wrapper over Python dictionaries representing a batch. It provides a
+    dictionary-like interface to access its fields. This class can be used in
+    the followed way
+
+    .. code-block:: python
+
+            hparams = {
+                'dataset': { 'files': 'data.txt', 'vocab_file': 'vocab.txt' },
+                'batch_size': 1
+            }
+
+            data = MonoTextData(hparams)
+            iterator = DataIterator(data)
+            model = BERTEncoder(pretrained_model_name="bert-base-uncased")
+
+            for batch in iterator:
+                # batch is Batch object and contains the following fields
+                # batch == {
+                #    'text': [['<BOS>', 'example', 'sequence', '<EOS>']],
+                #    'text_ids': [[1, 5, 10, 2]],
+                #    'length': [4]
+                # }
+
+                input_ids = torch.tensor(batch['text_ids'])
+                input_length = (1 - (input_ids == 0).int()).sum(dim=1)
+
+                bert_embeddings, _ = model(input_ids, input_length)
     """
 
     def __init__(self, batch_size: int, batch: Optional[Dict[str, Any]] = None,