add metric

README.md

@@ -1,7 +1,5 @@
 ---
 title: bash_eval
-datasets:
--  
 tags:
 - evaluate
 - metric

app.py

@@ -2,5 +2,5 @@ import evaluate
 from evaluate.utils import launch_gradio_widget
 
 
-module = evaluate.load("shuyanzh/bash_eval")
+module = evaluate.load("neulab/tldr_eval")
 launch_gradio_widget(module)

bash_eval.py

@@ -1,95 +0,0 @@
-# Copyright 2020 The HuggingFace Datasets Authors and the current dataset script contributor.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-"""TODO: Add a description here."""
-
-import evaluate
-import datasets
-
-
-# TODO: Add BibTeX citation
-_CITATION = """\
-@InProceedings{huggingface:module,
-title = {A great new module},
-authors={huggingface, Inc.},
-year={2020}
-}
-"""
-
-# TODO: Add description of the module here
-_DESCRIPTION = """\
-This new module is designed to solve this great ML task and is crafted with a lot of care.
-"""
-
-
-# TODO: Add description of the arguments of the module here
-_KWARGS_DESCRIPTION = """
-Calculates how good are predictions given some references, using certain scores
-Args:
-    predictions: list of predictions to score. Each predictions
-        should be a string with tokens separated by spaces.
-    references: list of reference for each prediction. Each
-        reference should be a string with tokens separated by spaces.
-Returns:
-    accuracy: description of the first score,
-    another_score: description of the second score,
-Examples:
-    Examples should be written in doctest format, and should illustrate how
-    to use the function.
-
-    >>> my_new_module = evaluate.load("my_new_module")
-    >>> results = my_new_module.compute(references=[0, 1], predictions=[0, 1])
-    >>> print(results)
-    {'accuracy': 1.0}
-"""
-
-# TODO: Define external resources urls if needed
-BAD_WORDS_URL = "http://url/to/external/resource/bad_words.txt"
-
-
-@evaluate.utils.file_utils.add_start_docstrings(_DESCRIPTION, _KWARGS_DESCRIPTION)
-class bash_eval(evaluate.Metric):
-    """TODO: Short description of my evaluation module."""
-
-    def _info(self):
-        # TODO: Specifies the evaluate.EvaluationModuleInfo object
-        return evaluate.MetricInfo(
-            # This is the description that will appear on the modules page.
-            module_type="metric",
-            description=_DESCRIPTION,
-            citation=_CITATION,
-            inputs_description=_KWARGS_DESCRIPTION,
-            # This defines the format of each prediction and reference
-            features=datasets.Features({
-                'predictions': datasets.Value('int64'),
-                'references': datasets.Value('int64'),
-            }),
-            # Homepage of the module for documentation
-            homepage="http://module.homepage",
-            # Additional links to the codebase or references
-            codebase_urls=["http://github.com/path/to/codebase/of/new_module"],
-            reference_urls=["http://path.to.reference.url/new_module"]
-        )
-
-    def _download_and_prepare(self, dl_manager):
-        """Optional: download external resources useful to compute the scores"""
-        # TODO: Download external resources if needed
-        pass
-
-    def _compute(self, predictions, references):
-        """Returns the scores"""
-        # TODO: Compute the different scores of the module
-        accuracy = sum(i == j for i, j in zip(predictions, references)) / len(predictions)
-        return {
-            "accuracy": accuracy,
-        }

requirements.txt

@@ -1 +1,2 @@
-git+https://github.com/huggingface/evaluate@main
+git+https://github.com/huggingface/evaluate@main
+sacrebleu

run.py

@@ -0,0 +1,31 @@
+import json
+from tldr_eval import TLDREval
+from collections import Counter
+        
+with open('test_results_test_same.json', 'r') as f:
+    data = []
+    for line in f:
+        item = json.loads(line)
+        data.append(item)
+        
+split_data = [[] for _ in range(10)]
+qid_counter = Counter()
+for item in data:
+    if item['question_id'] in ['9931', '7895', '3740', '8077', '4737', '7057', '9530']:
+        continue
+    split_idx = qid_counter[item['question_id']]
+    split_data[split_idx].append(item)
+    qid_counter[item['question_id']] += 1
+assert all([len(x) in [918, 1845, 0] for x in split_data])
+print([len(x) for x in split_data])
+    
+refs = []
+preds = []
+for item in split_data[0]:
+    refs.append(item['gold'].replace('\n', ""))
+    preds.append(item['clean_code'].replace('\n', ""))
+
+
+evaluator  = TLDREval()
+metrics = evaluator._compute(preds, refs)
+print(metrics)

tests.py

@@ -1,17 +0,0 @@
-test_cases = [
-    {
-        "predictions": [0, 0],
-        "references": [1, 1],
-        "result": {"metric_score": 0}
-    },
-    {
-        "predictions": [1, 1],
-        "references": [1, 1],
-        "result": {"metric_score": 1}
-    },
-    {
-        "predictions": [1, 0],
-        "references": [1, 1],
-        "result": {"metric_score": 0.5}
-    }
-]

tldr_eval.py

@@ -0,0 +1,183 @@
+# Copyright 2020 The HuggingFace Datasets Authors and the current dataset script contributor.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""TODO: Add a description here."""
+
+import evaluate
+import datasets
+from collections import defaultdict, Counter
+import re
+import numpy as np
+from sacrebleu.metrics import BLEU
+
+# TODO: Add BibTeX citation
+_CITATION = """\
+@article{zhou2022doccoder,
+  title={DocCoder: Generating Code by Retrieving and Reading Docs},
+  author={Zhou, Shuyan and Alon, Uri and Xu, Frank F and Jiang, Zhengbao and Neubig, Graham},
+  journal={arXiv preprint arXiv:2207.05987},
+  year={2022}
+}
+"""
+
+_DESCRIPTION = """\
+This metric is used to evaluate the quality of a generated bash script.
+"""
+
+
+_KWARGS_DESCRIPTION = """
+predictions: list of str. The predictions
+references: list of str. The references
+
+Return
+
+"""
+
+VAR_STR = "[[VAR]]"
+
+
+def clean_command(s):
+    s = s.replace("sudo", "").strip()
+    s =  s.replace("`", "").replace('"', "").replace("'", "")
+    #  '>', '|', '+'
+    s = s.replace("|", " ").replace(">", " ").replace("<", " ")
+    s = " ".join(s.split())
+    return s
+
+def anonymize_command(s):
+    s = s.replace("={", " {")
+    var_to_pc_holder = defaultdict(lambda: len(var_to_pc_holder))
+    for var in re.findall("{{(.*?)}}", s):
+        _ = var_to_pc_holder[var]
+    for var, id in var_to_pc_holder.items():
+        var_str = "{{%s}}" % var
+        s = s.replace(var_str, f"{VAR_STR}_{id}")
+    # s = re.sub("{{.*?}}", VAR_STR, s)
+    return s
+
+def clean_anonymize_command(s):
+    return anonymize_command(clean_command(s))
+
+
+def get_bag_of_words(cmd):
+    cmd = clean_anonymize_command(cmd)
+    tokens = cmd.strip().split()
+    return tokens
+
+def calc_template_matching(gold, pred):
+    ag = clean_anonymize_command(gold)
+    ap = clean_anonymize_command(pred)
+    m = {'template_matching': int(ag == ap)}
+    return m
+
+def token_prf(tok_gold, tok_pred, match_blank=False):
+    if match_blank and len(tok_gold) == 0: # do not generate anything
+        if len(tok_pred) == 0:
+            m = {'r': 1, 'p': 1, 'f1': 1}
+        else:
+            m = {'r': 0, 'p': 0, 'f1': 0}
+    else:
+        tok_gold_dict = Counter(tok_gold)
+        tok_pred_dict = Counter(tok_pred)
+        tokens = set([*tok_gold_dict] + [*tok_pred_dict])
+        hit = 0
+        for token in tokens:
+            hit += min(tok_gold_dict.get(token, 0), tok_pred_dict.get(token, 0))
+        p = hit / (sum(tok_pred_dict.values()) + 1e-10)
+        r = hit / (sum(tok_gold_dict.values()) + 1e-10)
+        f1 = 2 * p * r / (p + r + 1e-10)
+        m = {'r': r, 'p': p, 'f1': f1}
+    return m
+
+def measure_bag_of_word(gold, pred):
+    tok_gold = get_bag_of_words(gold)
+    tok_pred = get_bag_of_words(pred)
+    m = token_prf(tok_gold, tok_pred) # whole sentence
+    gold_cmd = tok_gold[0] if len(tok_gold) else "NONE_GOLD"
+    pred_cmd = tok_pred[0] if len(tok_pred) else "NONE_PRED"
+    m = {**m, 'command_accuracy': int(gold_cmd == pred_cmd)}
+
+    return m
+
+def tldr_metrics(references, predictions):
+    assert len(references) == len(predictions)
+    metric_list = defaultdict(list)
+    for ref, pred in zip(references, predictions):
+        for k, v in calc_template_matching(ref, pred).items():
+            metric_list[k].append(v)
+        for k, v in measure_bag_of_word(ref, pred).items():
+            metric_list[k].append(v)
+
+    for k, v in metric_list.items():
+        metric_list[k] = np.mean(v)
+
+    def clean_for_bleu(s):
+        s = s.replace("sudo", "").strip()
+        s = s.replace("`", "").replace('"', "").replace("'", "")
+        #  '>', '|', '+'
+        s = s.replace("|", " ").replace(">", " ").replace("<", " ")
+        s = " ".join(s.split())
+        s = s.replace("={", " {")
+        var_to_pc_holder = defaultdict(lambda: len(var_to_pc_holder))
+        for var in re.findall("{{(.*?)}}", s):
+            _ = var_to_pc_holder[var]
+        for var, id in var_to_pc_holder.items():
+            var_str = "{{%s}}" % var
+            s = s.replace(var_str, f"${id}")
+        # s = re.sub("{{.*?}}", VAR_STR, s)
+        # print(s)
+        return s
+    
+    def to_characters(s):
+        # s = s.replace(" ", "")
+        # s = ' '.join(list(s))
+        return s
+    # character level
+    bleu = BLEU(tokenize='char')
+    predictions = [to_characters(clean_for_bleu(x)) for x in predictions]
+    references = [to_characters(clean_for_bleu(x)) for x in references]
+    bleu_score = bleu.corpus_score(predictions, [references]).score
+    metric_list['bleu_char'] = bleu_score
+    return metric_list
+
+@evaluate.utils.file_utils.add_start_docstrings(_DESCRIPTION, _KWARGS_DESCRIPTION)
+class TLDREval(evaluate.Metric):
+    """Evaluate Bash scripts."""
+
+    def _info(self):
+        return evaluate.MetricInfo(
+            # This is the description that will appear on the modules page.
+            module_type="metric",
+            description=_DESCRIPTION,
+            citation=_CITATION,
+            inputs_description=_KWARGS_DESCRIPTION,
+            # This defines the format of each prediction and reference
+            features=datasets.Features({
+                "predictions": datasets.Value("string", id="sequence"),
+                "references": datasets.Value("string", id="sequence"),
+            }),
+            # Homepage of the module for documentation
+            homepage="https://github.com/shuyanzhou/docprompting",
+            # Additional links to the codebase or references
+            codebase_urls=["https://github.com/shuyanzhou/docprompting"],
+            reference_urls=["https://github.com/shuyanzhou/docprompting"]
+        )
+
+    def _compute(self, predictions, references):
+        """Returns the scores"""
+        metrics = tldr_metrics(references, predictions)
+        # rename for better display
+        metrics['token_recall'] = metrics.pop('r')
+        metrics['token_precision'] = metrics.pop('p')
+        metrics['token_f1'] = metrics.pop('f1')
+        return dict(metrics)