Merge pull request #686 from QData/updates

[Draft] Updating transformers, datasets, gensim

docs/conf.py

@@ -21,7 +21,7 @@
 author = "UVA QData Lab"
 
 # The full version, including alpha/beta/rc tags
-release = "0.3.6"
+release = "0.3.7"
 
 # Set master doc to `index.rst`.
 master_doc = "index"

requirements.txt

@@ -5,13 +5,13 @@ filelock
 language_tool_python
 lemminflect
 lru-dict
-datasets==2.2.2
+datasets==2.4.0
 nltk
 numpy>=1.21.0
 pandas>=1.0.1
 scipy>=1.4.1
 torch>=1.7.0,!=1.8
-transformers>=3.3.0
+transformers>=4.21.0
 terminaltables
 tqdm
 word2number

setup.py

@@ -35,11 +35,11 @@
 ]
 
 extras["optional"] = [
-    "sentence_transformers>0.2.6",
+    "sentence_transformers==2.2.0",
     "stanza",
     "visdom",
     "wandb",
-    "gensim==3.8.3",
+    "gensim==4.1.2",
 ]
 
 # For developers, install development tools along with all optional dependencies.

textattack/constraints/semantics/sentence_encoders/universal_sentence_encoder/multilingual_universal_sentence_encoder.py

@@ -2,12 +2,12 @@
 multilingual universal sentence encoder
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 """
-import tensorflow_text  # noqa: F401
 
 from textattack.constraints.semantics.sentence_encoders import SentenceEncoder
 from textattack.shared.utils import LazyLoader
 
 hub = LazyLoader("tensorflow_hub", globals(), "tensorflow_hub")
+tensorflow_text = LazyLoader("tensorflow_text", globals(), "tensorflow_text")
 
 
 class MultilingualUniversalSentenceEncoder(SentenceEncoder):

textattack/models/tokenizers/glove_tokenizer.py

@@ -46,7 +46,7 @@ def __init__(
         word_list_file = tempfile.NamedTemporaryFile()
         word_list_file.write(json.dumps(word_id_map).encode())
 
-        word_level = hf_tokenizers.models.WordLevel(
+        word_level = hf_tokenizers.models.WordLevel.from_file(
             word_list_file.name, unk_token=str(unk_token)
         )
         tokenizer = hf_tokenizers.Tokenizer(word_level)

textattack/shared/utils/tensor.py

@@ -9,6 +9,8 @@ def batch_model_predict(model_predict, inputs, batch_size=32):
     """
     outputs = []
     i = 0
+    # print("batch_model_predict", inputs.shape)
+    # print("inputs:", inputs)
     while i < len(inputs):
         batch = inputs[i : i + batch_size]
         batch_preds = model_predict(batch)

textattack/shared/word_embeddings.py

@@ -312,9 +312,7 @@ class GensimWordEmbedding(AbstractWordEmbedding):
     def __init__(self, keyed_vectors):
         gensim = utils.LazyLoader("gensim", globals(), "gensim")
 
-        if isinstance(
-            keyed_vectors, gensim.models.keyedvectors.WordEmbeddingsKeyedVectors
-        ):
+        if isinstance(keyed_vectors, gensim.models.KeyedVectors):
             self.keyed_vectors = keyed_vectors
         else:
             raise ValueError(
@@ -335,11 +333,11 @@ def __getitem__(self, index):
         """
         if isinstance(index, str):
             try:
-                index = self.keyed_vectors.vocab.get(index).index
+                index = self.keyed_vectors.key_to_index.get(index)
             except KeyError:
                 return None
         try:
-            return self.keyed_vectors.vectors_norm[index]
+            return self.keyed_vectors.get_normed_vectors()[index]
         except IndexError:
             # word embedding ID out of bounds
             return None
@@ -352,10 +350,10 @@ def word2index(self, word):
         Returns:
             index (int)
         """
-        vocab = self.keyed_vectors.vocab.get(word)
+        vocab = self.keyed_vectors.key_to_index.get(word)
         if vocab is None:
             raise KeyError(word)
-        return vocab.index
+        return vocab
 
     def index2word(self, index):
         """
@@ -368,7 +366,7 @@ def index2word(self, index):
         """
         try:
             # this is a list, so the error would be IndexError
-            return self.keyed_vectors.index2word[index]
+            return self.keyed_vectors.index_to_key[index]
         except IndexError:
             raise KeyError(index)
 
@@ -386,8 +384,8 @@ def get_mse_dist(self, a, b):
         try:
             mse_dist = self._mse_dist_mat[a][b]
         except KeyError:
-            e1 = self.keyed_vectors.vectors_norm[a]
-            e2 = self.keyed_vectors.vectors_norm[b]
+            e1 = self.keyed_vectors.get_normed_vectors()[a]
+            e2 = self.keyed_vectors.get_normed_vectors()[b]
             e1 = torch.tensor(e1).to(utils.device)
             e2 = torch.tensor(e2).to(utils.device)
             mse_dist = torch.sum((e1 - e2) ** 2).item()
@@ -406,9 +404,9 @@ def get_cos_sim(self, a, b):
             distance (float): cosine similarity
         """
         if not isinstance(a, str):
-            a = self.keyed_vectors.index2word[a]
+            a = self.keyed_vectors.index_to_key[a]
         if not isinstance(b, str):
-            b = self.keyed_vectors.index2word[b]
+            b = self.keyed_vectors.index_to_key[b]
         cos_sim = self.keyed_vectors.similarity(a, b)
         return cos_sim
 
@@ -421,7 +419,7 @@ def nearest_neighbours(self, index, topn, return_words=True):
         Returns:
             neighbours (list[int]): List of indices of the nearest neighbours
         """
-        word = self.keyed_vectors.index2word[index]
+        word = self.keyed_vectors.index_to_key[index]
         return [
             self.word2index(i[0])
             for i in self.keyed_vectors.similar_by_word(word, topn)