Update process.py

process.py

@@ -6,35 +6,38 @@ from tqdm import tqdm
 from typing import Dict
 from datasets import load_dataset
 
-SKIP_FEATURE = ["sentence_answer", "paragraph_answer", "paragraph_sentence"]
 
+SEP_TOKEN = "\n"
 
-def process_single_data(data: Dict):
-    """ Convert single raw json data into QG format """
-    example = {'question': data["Question"], 'paragraph': data["Tweet"], "paragraph_id": data['qid'], 'answer': data['Answer'][0]}
-    for i in SKIP_FEATURE:
-      example[i] = None
-    return example
+
+def create_data(hf_data):
+    df = hf_data.to_pandas()
+    output = []
+    for tweet, g in df.groupby("Tweet"):
+        example = {
+            'paragraph': tweet.replace(SEP_TOKEN, " "),
+            "paragraph_id": '-'.join(g['qid']),
+            'questions': [_g.replace(SEP_TOKEN, " ") for _g in g['Question']],
+            'answers': [_g[0].replace(SEP_TOKEN, " ") for _g in g['Answer']],
+        }
+        example["questions_answers"] = SEP_TOKEN.join([f"Q: {q}, A: {a}" for q, a in zip(example["questions"], example["answers"])])
+        output.append(example)
+     return output
 
 
 if __name__ == '__main__':
     tweet_qa = load_dataset("tweet_qa")
-    data_dev = tweet_qa['validation']
+    data_valid = create_data(tweet_qa['validation'])
+    data_train = create_data(tweet_qa['train'])
     seed(1)
-    test_context_len = 900
-
-    # create test set from training
-    data_train = tweet_qa['train']
-    context = sorted(list(set(data_train['Tweet'])))
-    shuffle(context)
-    data_test = [data_train[i] for i in range(len(data_train)) if data_train[i]['Tweet'] in context[:test_context_len]]
-    data_train = [data_train[i] for i in range(len(data_train)) if data_train[i]['Tweet'] in context[test_context_len:]]
-    print(f'train ({len(data_train)}, test ({len(data_test)}), dev ({len(data_dev)})')
-    data_all = {'train': data_train, 'validation': data_dev, 'test': data_test}
+    test_len = len(data_valid)
+    shuffle(data_train)
+    data_test = data_train[:test_len]
+    data_train = data_train[test_len:]
+    data_all = {'train': data_train, 'validation': data_valid, 'test': data_test}
     output = './data/processed'
     os.makedirs(output, exist_ok=True)
     for k, _data in data_all.items():
         with open('{}/{}.jsonl'.format(output, k), 'w') as f:
             for single_data in tqdm(_data):
-                single_data = process_single_data(single_data)
                 f.write(json.dumps(single_data) + '\n')