Clean up code

codebleu/codebleu.py

@@ -7,7 +7,6 @@
 from . import bleu, dataflow_match, syntax_match, weighted_ngram_match
 
 PACKAGE_DIR = Path(__file__).parent
-# AVAILABLE_LANGS = ['java', 'javascript', 'c_sharp', 'php', 'go', 'python', 'ruby']
 AVAILABLE_LANGS = ["java", "javascript", "c_sharp", "php", "c", "cpp", "python"]  # keywords available
 
 
@@ -56,7 +55,8 @@ def tokenizer(s):
     ngram_match_score = bleu.corpus_bleu(tokenized_refs, tokenized_hyps)
 
     # calculate weighted ngram match
-    keywords = [x.strip() for x in open(keywords_dir / (lang + ".txt"), "r", encoding="utf-8").readlines()]
+    with open(keywords_dir / (lang + ".txt"), "r", encoding="utf-8") as f:
+        keywords = [x.strip() for x in f.readlines()]
 
     def make_weights(reference_tokens, key_word_list):
         return {token: 1 if token in key_word_list else 0.2 for token in reference_tokens}
@@ -74,25 +74,14 @@ def make_weights(reference_tokens, key_word_list):
     # calculate dataflow match
     dataflow_match_score = dataflow_match.corpus_dataflow_match(references, hypothesis, lang, lang_so_file)
 
-    # print(
-    #     "ngram match: {0}, weighted ngram match: {1}, syntax_match: {2}, dataflow_match: {3}".format(
-    #         ngram_match_score,
-    #         weighted_ngram_match_score,
-    #         syntax_match_score,
-    #         dataflow_match_score,
-    #     )
-    # )
-
     alpha, beta, gamma, theta = weights
     code_bleu_score = (
         alpha * ngram_match_score
         + beta * weighted_ngram_match_score
         + gamma * syntax_match_score
-        + theta * (dataflow_match_score or 1)
+        + theta * (dataflow_match_score or 0)
     )
 
-    # print("CodeBLEU score: ", code_bleu_score)
-
     return {
         "codebleu": code_bleu_score,
         "ngram_match_score": ngram_match_score,

evaluate_app/codebleu.py

@@ -106,7 +106,7 @@ def _info(self):
     def _download_and_prepare(self, dl_manager):
         """Optional: download external resources useful to compute the scores"""
         # workarounds as this file have to be named codebleu (evaluate library requirement)
-        self.codebleu_package = importlib.import_module('codebleu')
+        self.codebleu_package = importlib.import_module("codebleu")
         pass
 
     def _compute(self, predictions, references, lang, weights=(0.25, 0.25, 0.25, 0.25), tokenizer=None):

tests/test_codebleu.py

@@ -7,7 +7,8 @@
 
 
 @pytest.mark.parametrize(['predictions', 'references', 'codebleu'], [
-    (['some rannnndom words in length more than 3'], ['def test ( ) :\n pass'], 0.25),  # 'cause data_flow is 0 and considered as 1
+    (['some rannnndom words in length more than 3'],
+     ['def test ( ) :\n pass'], 0.25),  # 'cause data_flow is 0 and considered as 1
     (['def bar ( y , x ) :\n    a = x * x\n    return a'], ['def foo ( x ) :\n    return x'], 0.4),
     (['def foo ( x ) :\n    return x * x'], ['def bar ( x ) :\n    return x'], 0.6),
     (['def bar ( x ) :\n    return x'], ['def foo ( x ) :\n    return x'], 0.8),
@@ -19,7 +20,7 @@ def test_simple_cases(predictions: List[Any], references: List[Any], codebleu: f
     assert result['codebleu'] == pytest.approx(codebleu, 0.1)
 
 
-@pytest.mark.parametrize(['lang'], [(l,) for l in AVAILABLE_LANGS])
+@pytest.mark.parametrize(['lang'], [(lang,) for lang in AVAILABLE_LANGS])
 def test_exact_match_works_for_all_langs(lang: str) -> None:
     predictions = references = ['some matching string a couple of times']
     assert calc_codebleu(references, predictions, lang)['codebleu'] == 1.0