File size: 6,980 Bytes
f5d0a0d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
"""
Build the combined UDD-1 multi-domain dataset (40,000 sentences).

Reads sentence files from all domains, assigns domain-specific sent_id prefixes,
and creates stratified train/dev/test splits.

Domain mapping:
  - sentences_vlc.txt  -> prefix: vlc-    (Legal)
  - sentences_uvn.txt  -> prefix: uvn-    (News)
  - sentences_uvw.txt  -> prefix: uvw-    (Wikipedia)
  - sentences_uvb.txt  -> prefix: uvb-f-  (Fiction), uvb-n- (Non-fiction)

Output:
  - sentences_train.txt  (91.4%)
  - sentences_dev.txt    (4.3%)
  - sentences_test.txt   (4.3%)

Each line format: sent_id\tsentence
"""

import random
from os.path import dirname, isfile, join


# Split ratios
TRAIN_RATIO = 0.914
DEV_RATIO = 0.043
TEST_RATIO = 0.043


def load_sentences_with_prefix(filepath, prefix):
    """Load sentences from a file and assign sent_id prefix.

    Returns list of (sent_id, sentence) tuples.
    """
    sentences = []
    with open(filepath, "r", encoding="utf-8") as f:
        for line in f:
            line = line.strip()
            if not line:
                continue
            parts = line.split("\t")
            # Format: idx\tsentence
            if len(parts) == 2:
                idx = parts[0]
                sentence = parts[1]
                sent_id = f"{prefix}{idx}"
                sentences.append((sent_id, sentence))
            # Format: idx\tsource\tsentence (sentences_uvb.txt)
            elif len(parts) >= 3:
                idx = parts[0]
                source = parts[1]
                sentence = parts[2]
                sent_id = f"{prefix}{idx}"
                sentences.append((sent_id, sentence, source))
    return sentences


def load_uvb_sentences(filepath):
    """Load UVB sentences and split by fiction/non-fiction with proper prefixes."""
    fiction = []
    non_fiction = []
    fiction_idx = 0
    non_fiction_idx = 0

    with open(filepath, "r", encoding="utf-8") as f:
        for line in f:
            line = line.strip()
            if not line:
                continue
            parts = line.split("\t")
            if len(parts) >= 3:
                source = parts[1]
                sentence = parts[2]
                if source == "fiction":
                    fiction_idx += 1
                    fiction.append((f"uvb-f-{fiction_idx}", sentence))
                else:
                    non_fiction_idx += 1
                    non_fiction.append((f"uvb-n-{non_fiction_idx}", sentence))

    return fiction, non_fiction


def stratified_split(domain_sentences, seed=42):
    """Create stratified train/dev/test split preserving domain proportions.

    Args:
        domain_sentences: dict of domain_name -> list of (sent_id, sentence)
        seed: random seed for reproducibility

    Returns:
        train, dev, test lists of (sent_id, sentence)
    """
    random.seed(seed)

    train = []
    dev = []
    test = []

    for domain_name, sentences in domain_sentences.items():
        # Shuffle within each domain
        shuffled = list(sentences)
        random.shuffle(shuffled)

        n = len(shuffled)
        n_dev = max(1, round(n * DEV_RATIO))
        n_test = max(1, round(n * TEST_RATIO))
        n_train = n - n_dev - n_test

        train.extend(shuffled[:n_train])
        dev.extend(shuffled[n_train:n_train + n_dev])
        test.extend(shuffled[n_train + n_dev:])

        print(f"  {domain_name}: {n_train} train / {n_dev} dev / {n_test} test (total: {n})")

    return train, dev, test


def save_split(sentences, filepath):
    """Save a list of (sent_id, sentence) to file."""
    with open(filepath, "w", encoding="utf-8") as f:
        for sent_id, sentence in sentences:
            f.write(f"{sent_id}\t{sentence}\n")


def main():
    base_dir = dirname(dirname(__file__))

    # Define source files and their prefixes
    sources = {
        "vlc": ("sentences_vlc.txt", "vlc-"),
        "uvn": ("sentences_uvn.txt", "uvn-"),
        "uvw": ("sentences_uvw.txt", "uvw-"),
    }

    # Load sentences from each domain
    domain_sentences = {}

    for domain, (filename, prefix) in sources.items():
        filepath = join(base_dir, filename)
        if not isfile(filepath):
            print(f"Warning: {filepath} not found, skipping {domain}")
            continue
        sents = load_sentences_with_prefix(filepath, prefix)
        # Extract just (sent_id, sentence) tuples
        domain_sentences[domain] = [(s[0], s[1]) for s in sents]
        print(f"Loaded {len(domain_sentences[domain])} sentences from {filename}")

    # Load UVB (books) with fiction/non-fiction split
    uvb_filepath = join(base_dir, "sentences_uvb.txt")
    if isfile(uvb_filepath):
        fiction, non_fiction = load_uvb_sentences(uvb_filepath)
        domain_sentences["uvb-fiction"] = fiction
        domain_sentences["uvb-nonfiction"] = non_fiction
        print(f"Loaded {len(fiction)} fiction + {len(non_fiction)} non-fiction sentences from sentences_uvb.txt")
    else:
        print(f"Warning: {uvb_filepath} not found, skipping books domain")

    # Report totals
    total = sum(len(v) for v in domain_sentences.values())
    print(f"\nTotal sentences across all domains: {total}")

    # Create stratified split
    print("\nCreating stratified train/dev/test split...")
    train, dev, test = stratified_split(domain_sentences)

    print(f"\nSplit sizes:")
    print(f"  Train: {len(train)} ({100*len(train)/total:.1f}%)")
    print(f"  Dev:   {len(dev)} ({100*len(dev)/total:.1f}%)")
    print(f"  Test:  {len(test)} ({100*len(test)/total:.1f}%)")
    print(f"  Total: {len(train) + len(dev) + len(test)}")

    # Save splits
    save_split(train, join(base_dir, "sentences_train.txt"))
    save_split(dev, join(base_dir, "sentences_dev.txt"))
    save_split(test, join(base_dir, "sentences_test.txt"))

    print(f"\nSaved to:")
    print(f"  {join(base_dir, 'sentences_train.txt')}")
    print(f"  {join(base_dir, 'sentences_dev.txt')}")
    print(f"  {join(base_dir, 'sentences_test.txt')}")

    # Print domain distribution per split
    print("\nDomain distribution per split:")
    for split_name, split_data in [("Train", train), ("Dev", dev), ("Test", test)]:
        domain_counts = {}
        for sent_id, _ in split_data:
            # Determine domain from sent_id prefix
            if sent_id.startswith("vlc-"):
                domain = "legal"
            elif sent_id.startswith("uvn-"):
                domain = "news"
            elif sent_id.startswith("uvw-"):
                domain = "wikipedia"
            elif sent_id.startswith("uvb-f-"):
                domain = "fiction"
            elif sent_id.startswith("uvb-n-"):
                domain = "non-fiction"
            else:
                domain = "unknown"
            domain_counts[domain] = domain_counts.get(domain, 0) + 1

        counts_str = ", ".join(f"{d}: {c}" for d, c in sorted(domain_counts.items()))
        print(f"  {split_name}: {counts_str}")


if __name__ == "__main__":
    main()