Updated classification model tokenization logic. Added deberta, mpnet…

…, squeezenet for classification

CHANGELOG.md

@@ -4,6 +4,20 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.60.0] - 2021-02-02
+
+# Added
+
+- Added class weights support for Longformer classification
+- Added new classification models:
+  - SqueezeBert
+  - DeBERTa
+  - MPNet
+
+# Changed
+
+- Updated ClassificationModel logic to make it easier to add new models
+
 ## [0.51.16] - 2021-01-29
 
 ## Fixed
@@ -1386,7 +1400,11 @@ Model checkpoint is now saved for all epochs again.
 
 - This CHANGELOG file to hopefully serve as an evolving example of a standardized open source project CHANGELOG.
 
-[0.51.15]: https://github.com/ThilinaRajapakse/simpletransformers/compare/2af55e9...HEAD
+[0.60.0]: https://github.com/ThilinaRajapakse/simpletransformers/compare/5840749...HEAD
+
+[0.51.16]: https://github.com/ThilinaRajapakse/simpletransformers/compare/b42898e...5840749
+
+[0.51.15]: https://github.com/ThilinaRajapakse/simpletransformers/compare/2af55e9...b42898e
 
 [0.51.14]: https://github.com/ThilinaRajapakse/simpletransformers/compare/278fca1...2af55e9
 

docs/_docs/04-classification-specifics.md

@@ -2,7 +2,7 @@
 title: Classification Specifics
 permalink: /docs/classification-specifics/
 excerpt: "Specific notes for text classification tasks."
-last_modified_at: 2020/12/21 22:13:56
+last_modified_at: 2021/02/02 02:03:09
 toc: true
 ---
 
@@ -32,22 +32,25 @@ The process of performing text classification in Simple Transformers does not de
 
 New model types are regularly added to the library. Text classification tasks currently supports the model types given below.
 
-| Model       | Model code for `ClassificationModel` |
-| ----------- | ------------------------------------ |
-| ALBERT      | albert                               |
-| BERT        | bert                                 |
-| BERTweet    | bertweet                             |
-| CamemBERT   | camembert                            |
-| RoBERTa     | roberta                              |
-| DistilBERT  | distilbert                           |
-| ELECTRA     | electra                              |
-| FlauBERT    | flaubert                             |
-| *LayoutLM   | layoutlm                             |
-| Longformer  | longformer                           |
-| *MobileBERT | mobilebert                           |
-| XLM         | xlm                                  |
-| XLM-RoBERTa | xlmroberta                           |
-| XLNet       | xlnet                                |
+| Model        | Model code for `ClassificationModel` |
+| ------------ | ------------------------------------ |
+| ALBERT       | albert                               |
+| BERT         | bert                                 |
+| BERTweet     | bertweet                             |
+| CamemBERT    | camembert                            |
+| *DeBERTa     | deberta                              |
+| DistilBERT   | distilbert                           |
+| ELECTRA      | electra                              |
+| FlauBERT     | flaubert                             |
+| LayoutLM     | layoutlm                             |
+| *Longformer  | longformer                           |
+| *MPNet       | mpnet                                |
+| MobileBERT   | mobilebert                           |
+| RoBERTa      | roberta                              |
+| *SqueezeBert | squeezebert                          |
+| XLM          | xlm                                  |
+| XLM-RoBERTa  | xlmroberta                           |
+| XLNet        | xlnet                                |
 
 \* *Not available with Multi-label classification*
 

examples/t5/mt5_translation/test.py

@@ -33,4 +33,4 @@
 english_preds = model.predict(to_english)
 
 sin_eng_bleu = sacrebleu.corpus_bleu(english_preds, english_truth)
-print("Sinhalese to English: ", sin_eng_bleu.score)
+print("Sinhalese to English: ", sin_eng_bleu.score)

examples/t5/training_on_a_new_task/train.py

@@ -22,6 +22,6 @@
     "wandb_project": "Question Generation with T5",
 }
 
-model = T5Model("t5","t5-large",args=model_args)
+model = T5Model("t5", "t5-large", args=model_args)
 
 model.train_model(train_df, eval_data=eval_df)

setup.py

@@ -5,7 +5,7 @@
 
 setup(
     name="simpletransformers",
-    version="0.51.16",
+    version="0.60.0",
     author="Thilina Rajapakse",
     author_email="[email protected]",
     description="An easy-to-use wrapper library for the Transformers library.",

simpletransformers/classification/classification_model.py

@@ -27,7 +27,7 @@
     mean_squared_error,
     roc_curve,
     auc,
-    average_precision_score
+    average_precision_score,
 )
 from tensorboardX import SummaryWriter
 from torch.utils.data import DataLoader, RandomSampler, SequentialSampler, TensorDataset
@@ -46,11 +46,17 @@
 from transformers import (
     AlbertConfig,
     AlbertTokenizer,
+    AutoConfig,
+    AutoModelForSequenceClassification,
+    AutoTokenizer,
     BertConfig,
     BertTokenizer,
     BertweetTokenizer,
     CamembertConfig,
     CamembertTokenizer,
+    DebertaConfig,
+    DebertaForSequenceClassification,
+    DebertaTokenizer,
     DistilBertConfig,
     DistilBertTokenizer,
     ElectraConfig,
@@ -60,15 +66,17 @@
     LayoutLMConfig,
     LayoutLMTokenizer,
     LongformerConfig,
-    LongformerForSequenceClassification,
     LongformerTokenizer,
+    MPNetConfig,
+    MPNetForSequenceClassification,
+    MPNetTokenizer,
     MobileBertConfig,
-    MobileBertForSequenceClassification,
     MobileBertTokenizer,
-    ReformerConfig,
-    ReformerTokenizer,
     RobertaConfig,
     RobertaTokenizer,
+    SqueezeBertConfig,
+    SqueezeBertForSequenceClassification,
+    SqueezeBertTokenizer,
     WEIGHTS_NAME,
     XLMConfig,
     XLMRobertaConfig,
@@ -91,6 +99,8 @@
 from simpletransformers.classification.transformer_models.distilbert_model import DistilBertForSequenceClassification
 from simpletransformers.classification.transformer_models.flaubert_model import FlaubertForSequenceClassification
 from simpletransformers.classification.transformer_models.layoutlm_model import LayoutLMForSequenceClassification
+from simpletransformers.classification.transformer_models.longformer_model import LongformerForSequenceClassification
+from simpletransformers.classification.transformer_models.mobilebert_model import MobileBertForSequenceClassification
 from simpletransformers.classification.transformer_models.roberta_model import RobertaForSequenceClassification
 from simpletransformers.classification.transformer_models.xlm_model import XLMForSequenceClassification
 from simpletransformers.classification.transformer_models.xlm_roberta_model import XLMRobertaForSequenceClassification
@@ -100,7 +110,6 @@
 from simpletransformers.config.utils import sweep_config_to_sweep_values
 from simpletransformers.custom_models.models import ElectraForSequenceClassification
 
-from transformers.models.reformer import ReformerForSequenceClassification
 
 try:
     import wandb
@@ -112,11 +121,22 @@
 logger = logging.getLogger(__name__)
 
 
+MODELS_WITHOUT_CLASS_WEIGHTS_SUPPORT = ["squeezebert", "deberta", "mpnet"]
+
+MODELS_WITH_EXTRA_SEP_TOKEN = ["roberta", "camembert", "xlmroberta", "longformer", "mpnet"]
+
+MODELS_WITH_ADD_PREFIX_SPACE = ["roberta", "camembert", "xlmroberta", "longformer", "mpnet"]
+
+MODELS_WITHOUT_SLIDING_WINDOW_SUPPORT = ["squeezebert"]
+
+
 class ClassificationModel:
     def __init__(
         self,
         model_type,
         model_name,
+        tokenizer_type=None,
+        tokenizer_name=None,
         num_labels=None,
         weight=None,
         args=None,
@@ -132,6 +152,9 @@ def __init__(
         Args:
             model_type: The type of model (bert, xlnet, xlm, roberta, distilbert)
             model_name: The exact architecture and trained weights to use. This may be a Hugging Face Transformers compatible pre-trained model, a community model, or the path to a directory containing model files.
+            tokenizer_type: The type of tokenizer (auto, bert, xlnet, xlm, roberta, distilbert, etc.) to use. If a string is passed, Simple Transformers will try to initialize a tokenizer class from the available MODEL_CLASSES.
+                                Alternatively, a Tokenizer class (subclassed from PreTrainedTokenizer) can be passed.
+            tokenizer_name: The name/path to the tokenizer. If the tokenizer_type is not specified, the model_type will be used to determine the type of the tokenizer.
             num_labels (optional): The number of labels or classes in the dataset.
             weight (optional): A list of length num_labels containing the weights to assign to each label for loss calculation.
             args (optional): Default args will be used if this parameter is not provided. If provided, it should be a dict containing the args that should be changed in the default args.
@@ -143,17 +166,20 @@ def __init__(
 
         MODEL_CLASSES = {
             "albert": (AlbertConfig, AlbertForSequenceClassification, AlbertTokenizer),
+            "auto": (AutoConfig, AutoModelForSequenceClassification, AutoTokenizer),
             "bert": (BertConfig, BertForSequenceClassification, BertTokenizer),
             "bertweet": (RobertaConfig, RobertaForSequenceClassification, BertweetTokenizer),
             "camembert": (CamembertConfig, CamembertForSequenceClassification, CamembertTokenizer),
+            "deberta": (DebertaConfig, DebertaForSequenceClassification, DebertaTokenizer),
             "distilbert": (DistilBertConfig, DistilBertForSequenceClassification, DistilBertTokenizer),
             "electra": (ElectraConfig, ElectraForSequenceClassification, ElectraTokenizer),
             "flaubert": (FlaubertConfig, FlaubertForSequenceClassification, FlaubertTokenizer),
             "layoutlm": (LayoutLMConfig, LayoutLMForSequenceClassification, LayoutLMTokenizer),
             "longformer": (LongformerConfig, LongformerForSequenceClassification, LongformerTokenizer),
             "mobilebert": (MobileBertConfig, MobileBertForSequenceClassification, MobileBertTokenizer),
-            "reformer": (ReformerConfig, ReformerForSequenceClassification, ReformerTokenizer),
+            "mpnet": (MPNetConfig, MPNetForSequenceClassification, MPNetTokenizer),
             "roberta": (RobertaConfig, RobertaForSequenceClassification, RobertaTokenizer),
+            "squeezebert": (SqueezeBertConfig, SqueezeBertForSequenceClassification, SqueezeBertTokenizer),
             "xlm": (XLMConfig, XLMForSequenceClassification, XLMTokenizer),
             "xlmroberta": (XLMRobertaConfig, XLMRobertaForSequenceClassification, XLMRobertaTokenizer),
             "xlnet": (XLNetConfig, XLNetForSequenceClassification, XLNetTokenizer),
@@ -166,6 +192,9 @@ def __init__(
         elif isinstance(args, ClassificationArgs):
             self.args = args
 
+        if model_type in MODELS_WITHOUT_SLIDING_WINDOW_SUPPORT and self.args.sliding_window:
+            raise ValueError("{} does not currently support sliding window".format(model_type))
+
         if self.args.thread_count:
             torch.set_num_threads(self.args.thread_count)
 
@@ -200,13 +229,24 @@ def __init__(
             self.args.labels_list = [i for i in range(len_labels_list)]
 
         config_class, model_class, tokenizer_class = MODEL_CLASSES[model_type]
+
+        if tokenizer_type is not None:
+            if isinstance(tokenizer_type, str):
+                _, _, tokenizer_class = MODEL_CLASSES[tokenizer_type]
+            else:
+                tokenizer_class = tokenizer_type
+
         if num_labels:
             self.config = config_class.from_pretrained(model_name, num_labels=num_labels, **self.args.config)
             self.num_labels = num_labels
         else:
             self.config = config_class.from_pretrained(model_name, **self.args.config)
             self.num_labels = self.config.num_labels
-        self.weight = weight
+
+        if model_type in MODELS_WITHOUT_CLASS_WEIGHTS_SUPPORT and weight is not None:
+            raise ValueError("{} does not currently support class weights".format(model_type))
+        else:
+            self.weight = weight
 
         if use_cuda:
             if torch.cuda.is_available():
@@ -275,17 +315,20 @@ def __init__(
             except AttributeError:
                 raise AttributeError("fp16 requires Pytorch >= 1.6. Please update Pytorch or turn off fp16.")
 
-        if model_name in [
+        if tokenizer_name is None:
+            tokenizer_name = model_name
+
+        if tokenizer_name in [
             "vinai/bertweet-base",
             "vinai/bertweet-covid19-base-cased",
             "vinai/bertweet-covid19-base-uncased",
         ]:
             self.tokenizer = tokenizer_class.from_pretrained(
-                model_name, do_lower_case=self.args.do_lower_case, normalization=True, **kwargs
+                tokenizer_name, do_lower_case=self.args.do_lower_case, normalization=True, **kwargs
             )
         else:
             self.tokenizer = tokenizer_class.from_pretrained(
-                model_name, do_lower_case=self.args.do_lower_case, **kwargs
+                tokenizer_name, do_lower_case=self.args.do_lower_case, **kwargs
             )
 
         if self.args.special_tokens_list:
@@ -294,6 +337,8 @@ def __init__(
 
         self.args.model_name = model_name
         self.args.model_type = model_type
+        self.args.tokenizer_name = tokenizer_name
+        self.args.tokenizer_type = tokenizer_type
 
         if model_type in ["camembert", "xlmroberta"]:
             warnings.warn(
@@ -1184,7 +1229,7 @@ def load_and_cache_examples(
                     sep_token=tokenizer.sep_token,
                     # RoBERTa uses an extra separator b/w pairs of sentences,
                     # cf. github.com/pytorch/fairseq/commit/1684e166e3da03f5b600dbb7855cb98ddfcd0805
-                    sep_token_extra=bool(args.model_type in ["roberta", "camembert", "xlmroberta", "longformer"]),
+                    sep_token_extra=args.model_type in MODELS_WITH_EXTRA_SEP_TOKEN,
                     # PAD on the left for XLNet
                     pad_on_left=bool(args.model_type in ["xlnet"]),
                     pad_token=tokenizer.convert_tokens_to_ids([tokenizer.pad_token])[0],
@@ -1196,7 +1241,7 @@ def load_and_cache_examples(
                     sliding_window=args.sliding_window,
                     flatten=not evaluate,
                     stride=args.stride,
-                    add_prefix_space=bool(args.model_type in ["roberta", "camembert", "xlmroberta", "longformer"]),
+                    add_prefix_space=args.model_type in MODELS_WITH_ADD_PREFIX_SPACE,
                     # avoid padding in case of single example/online inferencing to decrease execution time
                     pad_to_max_length=bool(len(examples) > 1),
                     args=args,
@@ -1236,8 +1281,10 @@ def load_and_cache_examples(
             else:
                 return dataset
         else:
-            train_dataset = ClassificationDataset(examples, self.tokenizer, self.args, mode=mode, multi_label=multi_label, output_mode=output_mode)
-            return train_dataset
+            dataset = ClassificationDataset(
+                examples, self.tokenizer, self.args, mode=mode, multi_label=multi_label, output_mode=output_mode
+            )
+            return dataset
 
     def compute_metrics(self, preds, model_outputs, labels, eval_examples=None, multi_label=False, **kwargs):
         """
@@ -1302,7 +1349,10 @@ def compute_metrics(self, preds, model_outputs, labels, eval_examples=None, mult
                 auroc = auc(fpr, tpr)
                 auprc = average_precision_score(labels, scores)
                 return (
-                    {**{"mcc": mcc, "tp": tp, "tn": tn, "fp": fp, "fn": fn, "auroc": auroc, "auprc": auprc}, **extra_metrics},
+                    {
+                        **{"mcc": mcc, "tp": tp, "tn": tn, "fp": fp, "fn": fn, "auroc": auroc, "auprc": auprc},
+                        **extra_metrics,
+                    },
                     wrong,
                 )
         else:
@@ -1575,7 +1625,7 @@ def _move_model_to_device(self):
 
     def _get_inputs_dict(self, batch):
         if isinstance(batch[0], dict):
-            inputs = {key: value.squeeze().to(self.device) for key, value in batch[0].items()}
+            inputs = {key: value.squeeze(1).to(self.device) for key, value in batch[0].items()}
             inputs["labels"] = batch[1].to(self.device)
         else:
             batch = tuple(t.to(self.device) for t in batch)

simpletransformers/classification/classification_utils.py

@@ -96,15 +96,15 @@ def preprocess_data(data):
             max_length=args.max_seq_length,
             truncation=True,
             padding="max_length",
-            return_tensors="pt"
+            return_tensors="pt",
         )
     else:
         tokenized_example = tokenizer.encode_plus(
             text=example.text_a,
             max_length=args.max_seq_length,
             truncation=True,
             padding="max_length",
-            return_tensors="pt"
+            return_tensors="pt",
         )
 
     return {**tokenized_example, "label": example.label}
@@ -600,7 +600,7 @@ def __init__(
             self.data = [
                 dict(
                     json.load(open(os.path.join(data_path, l + self.data_type_extension))),
-                    **{"images": l + image_type_extension}
+                    **{"images": l + image_type_extension},
                 )
                 for l in files_list
             ]