Merge pull request #644 from QData/fixes-and-improvements

Bug fixes and tweaks

Author: jxmorris12

.github/workflows/check-formatting.yml

@@ -26,9 +26,9 @@ jobs:
     - name: Install dependencies
       run: |
         python -m pip install --upgrade pip setuptools wheel
-        pip install black flake8 isort # Testing packages
         python setup.py install_egg_info # Workaround https://github.com/pypa/pip/issues/4537
         pip install -e .[dev]
+        pip install black flake8 isort --upgrade # Testing packages
     - name: Check code format with black and isort
       run: |
         make lint

CONTRIBUTING.md

@@ -118,7 +118,7 @@ Follow these steps to start contributing:
    ```bash
    $ cd TextAttack
    $ pip install -e . ".[dev]"
-   $ pip install black isort pytest pytest-xdist
+   $ pip install black docformatter isort pytest pytest-xdist
    ```
 
    This will install `textattack` in editable mode and install `black` and 

docs/conf.py

@@ -21,7 +21,7 @@
 author = "UVA QData Lab"
 
 # The full version, including alpha/beta/rc tags
-release = "0.3.4"
+release = "0.3.5"
 
 # Set master doc to `index.rst`.
 master_doc = "index"

tests/test_attacked_text.py

@@ -70,7 +70,7 @@ def test_window_around_index(self, attacked_text):
 
     def test_big_window_around_index(self, attacked_text):
         assert (
-            attacked_text.text_window_around_index(0, 10 ** 5) + "."
+            attacked_text.text_window_around_index(0, 10**5) + "."
         ) == attacked_text.text
 
     def test_window_around_index_start(self, attacked_text):

tests/test_word_embedding.py

@@ -10,7 +10,7 @@ def test_embedding_paragramcf():
     word_embedding = WordEmbedding.counterfitted_GLOVE_embedding()
     assert pytest.approx(word_embedding[0][0]) == -0.022007
     assert pytest.approx(word_embedding["fawn"][0]) == -0.022007
-    assert word_embedding[10 ** 9] is None
+    assert word_embedding[10**9] is None
 
 
 def test_embedding_gensim():
@@ -37,7 +37,7 @@ def test_embedding_gensim():
     word_embedding = GensimWordEmbedding(keyed_vectors)
     assert pytest.approx(word_embedding[0][0]) == 1
     assert pytest.approx(word_embedding["bye-bye"][0]) == -1 / np.sqrt(2)
-    assert word_embedding[10 ** 9] is None
+    assert word_embedding[10**9] is None
 
     # test query functionality
     assert pytest.approx(word_embedding.get_cos_sim(1, 3)) == 0

textattack/attack.py

@@ -81,8 +81,8 @@ def __init__(
         constraints: List[Union[Constraint, PreTransformationConstraint]],
         transformation: Transformation,
         search_method: SearchMethod,
-        transformation_cache_size=2 ** 15,
-        constraint_cache_size=2 ** 15,
+        transformation_cache_size=2**15,
+        constraint_cache_size=2**15,
     ):
         """Initialize an attack object.
 
@@ -371,22 +371,23 @@ def _attack(self, initial_result):
         final_result = self.search_method(initial_result)
         self.clear_cache()
         if final_result.goal_status == GoalFunctionResultStatus.SUCCEEDED:
-            return SuccessfulAttackResult(
+            result = SuccessfulAttackResult(
                 initial_result,
                 final_result,
             )
         elif final_result.goal_status == GoalFunctionResultStatus.SEARCHING:
-            return FailedAttackResult(
+            result = FailedAttackResult(
                 initial_result,
                 final_result,
             )
         elif final_result.goal_status == GoalFunctionResultStatus.MAXIMIZING:
-            return MaximizedAttackResult(
+            result = MaximizedAttackResult(
                 initial_result,
                 final_result,
             )
         else:
             raise ValueError(f"Unrecognized goal status {final_result.goal_status}")
+        return result
 
     def attack(self, example, ground_truth_output):
         """Attack a single example.

textattack/attack_args.py

@@ -478,8 +478,8 @@ class _CommandLineAttackArgs:
     interactive: bool = False
     parallel: bool = False
     model_batch_size: int = 32
-    model_cache_size: int = 2 ** 18
-    constraint_cache_size: int = 2 ** 18
+    model_cache_size: int = 2**18
+    constraint_cache_size: int = 2**18
 
     @classmethod
     def _add_parser_args(cls, parser):

textattack/constraints/grammaticality/cola.py

@@ -43,7 +43,7 @@ def __init__(
 
         self.max_diff = max_diff
         self.model_name = model_name
-        self._reference_score_cache = lru.LRU(2 ** 10)
+        self._reference_score_cache = lru.LRU(2**10)
         model = AutoModelForSequenceClassification.from_pretrained(model_name)
         tokenizer = AutoTokenizer.from_pretrained(model_name)
         self.model = HuggingFaceModelWrapper(model, tokenizer)

textattack/constraints/grammaticality/language_models/google_language_model/alzantot_goog_lm.py

@@ -49,7 +49,7 @@ def __init__(self):
                 self.sess, self.graph, self.PBTXT_PATH, self.CKPT_PATH
             )
 
-        self.lm_cache = lru.LRU(2 ** 18)
+        self.lm_cache = lru.LRU(2**18)
 
     def clear_cache(self):
         self.lm_cache.clear()

textattack/constraints/grammaticality/part_of_speech.py

@@ -56,7 +56,7 @@ def __init__(
         self.language_nltk = language_nltk
         self.language_stanza = language_stanza
 
-        self._pos_tag_cache = lru.LRU(2 ** 14)
+        self._pos_tag_cache = lru.LRU(2**14)
         if tagger_type == "flair":
             if tagset == "universal":
                 self._flair_pos_tagger = SequenceTagger.load("upos-fast")
@@ -93,7 +93,8 @@ def _get_pos(self, before_ctx, word, after_ctx):
 
             if self.tagger_type == "flair":
                 context_key_sentence = Sentence(
-                    context_key, use_tokenizer=textattack.shared.utils.words_from_text
+                    context_key,
+                    use_tokenizer=textattack.shared.utils.TextAttackFlairTokenizer(),
                 )
                 self._flair_pos_tagger.predict(context_key_sentence)
                 word_list, pos_list = textattack.shared.utils.zip_flair_result(