File size: 9,550 Bytes
65057ee
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
"""
data/dataset.py
===============
Phase 1: PyTorch DataLoader with Sliding-Window Chunking

Provides ``StressDataset`` — a ``torch.utils.data.Dataset`` that splits
long texts into overlapping chunks of ``chunk_size`` tokens with a
configurable ``stride``, preventing truncation loss for long Reddit posts.

Each chunk is treated as an independent sample during training / inference,
and results can be aggregated per-document at evaluation time via
``doc_index``.

Usage
-----
>>> from data.dataset import StressDataset, create_dataloaders
>>> dataset = StressDataset(texts, labels, domains)
>>> train_dl, val_dl, test_dl = create_dataloaders(texts, labels, domains)
"""

from __future__ import annotations

from typing import Optional

import torch
from torch.utils.data import DataLoader, Dataset


# ---------------------------------------------------------------------------
# Vocabulary builder (simple word-level tokenizer)
# ---------------------------------------------------------------------------

_PAD_TOKEN = "<PAD>"
_UNK_TOKEN = "<UNK>"


class SimpleVocab:
    """Minimal word-level vocabulary for the CNN model.

    Assigns a unique integer to each token seen during ``build()``.
    """

    def __init__(self) -> None:
        self.token2idx: dict[str, int] = {_PAD_TOKEN: 0, _UNK_TOKEN: 1}
        self.idx2token: dict[int, str] = {0: _PAD_TOKEN, 1: _UNK_TOKEN}
        self.pad_idx: int = 0
        self.unk_idx: int = 1

    def build(self, texts: list[str], min_freq: int = 2) -> "SimpleVocab":
        """Build vocabulary from a list of texts.

        Parameters
        ----------
        texts : list[str]
            Raw text strings.
        min_freq : int
            Minimum token frequency to be included.

        Returns
        -------
        SimpleVocab
            self, for method chaining.
        """
        freq: dict[str, int] = {}
        for text in texts:
            for token in text.lower().split():
                freq[token] = freq.get(token, 0) + 1

        for token, count in freq.items():
            if count >= min_freq and token not in self.token2idx:
                idx = len(self.token2idx)
                self.token2idx[token] = idx
                self.idx2token[idx] = token

        return self

    def encode(self, text: str) -> list[int]:
        """Convert a text string to a list of token indices."""
        return [
            self.token2idx.get(t, self.unk_idx) for t in text.lower().split()
        ]

    def __len__(self) -> int:
        return len(self.token2idx)


# ---------------------------------------------------------------------------
# Sliding-Window Chunking Dataset
# ---------------------------------------------------------------------------

DEFAULT_CHUNK_SIZE: int = 200
DEFAULT_STRIDE: int = 50


class StressDataset(Dataset):
    """PyTorch Dataset with sliding-window chunking for long texts.

    Parameters
    ----------
    texts : list[str]
        Raw text strings.
    labels : list[int]
        Binary labels (0 = no stress, 1 = stress).
    domains : list[str]
        Domain tags (e.g. ``'reddit_long'``, ``'twitter_short'``).
    vocab : SimpleVocab, optional
        Pre-built vocabulary. If ``None``, one is built from ``texts``.
    chunk_size : int
        Maximum number of tokens per chunk.
    stride : int
        Step size between consecutive chunks.
    """

    def __init__(
        self,
        texts: list[str],
        labels: list[int],
        domains: list[str],
        vocab: SimpleVocab | None = None,
        chunk_size: int = DEFAULT_CHUNK_SIZE,
        stride: int = DEFAULT_STRIDE,
    ) -> None:
        if not (len(texts) == len(labels) == len(domains)):
            raise ValueError(
                "texts, labels, and domains must have the same length"
            )

        self.chunk_size = chunk_size
        self.stride = stride

        # Build or reuse vocabulary
        if vocab is None:
            self.vocab = SimpleVocab().build(texts)
        else:
            self.vocab = vocab

        # Pre-compute all chunks
        self._chunks: list[torch.Tensor] = []
        self._labels: list[int] = []
        self._domains: list[str] = []
        self._doc_indices: list[int] = []  # maps chunk → original doc

        for doc_idx, (text, label, domain) in enumerate(
            zip(texts, labels, domains)
        ):
            token_ids = self.vocab.encode(text)

            if len(token_ids) == 0:
                # Empty text → single padded chunk
                chunk = torch.zeros(chunk_size, dtype=torch.long)
                self._chunks.append(chunk)
                self._labels.append(label)
                self._domains.append(domain)
                self._doc_indices.append(doc_idx)
                continue

            # Generate sliding-window chunks
            chunks_created = 0
            for start in range(0, len(token_ids), stride):
                end = start + chunk_size
                chunk_ids = token_ids[start:end]

                # Pad if shorter than chunk_size
                if len(chunk_ids) < chunk_size:
                    chunk_ids = chunk_ids + [self.vocab.pad_idx] * (
                        chunk_size - len(chunk_ids)
                    )

                self._chunks.append(torch.tensor(chunk_ids, dtype=torch.long))
                self._labels.append(label)
                self._domains.append(domain)
                self._doc_indices.append(doc_idx)
                chunks_created += 1

                # Stop if we've consumed the entire text
                if end >= len(token_ids):
                    break

    def __len__(self) -> int:
        return len(self._chunks)

    def __getitem__(self, idx: int) -> dict[str, torch.Tensor | int | str]:
        return {
            "input_ids": self._chunks[idx],
            "label": self._labels[idx],
            "domain": self._domains[idx],
            "doc_index": self._doc_indices[idx],
        }


# ---------------------------------------------------------------------------
# DataLoader factory
# ---------------------------------------------------------------------------


def collate_fn(batch: list[dict]) -> dict[str, torch.Tensor | list]:
    """Custom collate function for ``StressDataset``.

    Stacks ``input_ids`` and ``label`` into tensors; keeps ``domain``
    and ``doc_index`` as lists.
    """
    input_ids = torch.stack([item["input_ids"] for item in batch])
    labels = torch.tensor([item["label"] for item in batch], dtype=torch.long)
    domains = [item["domain"] for item in batch]
    doc_indices = [item["doc_index"] for item in batch]

    return {
        "input_ids": input_ids,
        "labels": labels,
        "domains": domains,
        "doc_indices": doc_indices,
    }


def create_dataloaders(
    texts: list[str],
    labels: list[int],
    domains: list[str],
    vocab: SimpleVocab | None = None,
    chunk_size: int = DEFAULT_CHUNK_SIZE,
    stride: int = DEFAULT_STRIDE,
    batch_size: int = 32,
    train_ratio: float = 0.8,
    val_ratio: float = 0.1,
    seed: int = 42,
) -> tuple[DataLoader, DataLoader, DataLoader, SimpleVocab]:
    """Create train / validation / test DataLoaders.

    Parameters
    ----------
    texts, labels, domains : list
        Raw data arrays.
    vocab : SimpleVocab, optional
        Pre-built vocabulary; built from training split if ``None``.
    chunk_size, stride : int
        Sliding-window parameters.
    batch_size : int
        Batch size for all loaders.
    train_ratio, val_ratio : float
        Proportions for the train and validation splits.
        Test ratio = ``1 - train_ratio - val_ratio``.
    seed : int
        Random seed for reproducibility.

    Returns
    -------
    tuple[DataLoader, DataLoader, DataLoader, SimpleVocab]
        ``(train_loader, val_loader, test_loader, vocab)``
    """
    import random

    n = len(texts)
    indices = list(range(n))
    random.seed(seed)
    random.shuffle(indices)

    n_train = int(n * train_ratio)
    n_val = int(n * val_ratio)

    train_idx = indices[:n_train]
    val_idx = indices[n_train : n_train + n_val]
    test_idx = indices[n_train + n_val :]

    def _select(idx_list: list[int]) -> tuple[list[str], list[int], list[str]]:
        return (
            [texts[i] for i in idx_list],
            [labels[i] for i in idx_list],
            [domains[i] for i in idx_list],
        )

    train_texts, train_labels, train_domains = _select(train_idx)
    val_texts, val_labels, val_domains = _select(val_idx)
    test_texts, test_labels, test_domains = _select(test_idx)

    # Build vocab from training data only
    if vocab is None:
        vocab = SimpleVocab().build(train_texts)

    train_ds = StressDataset(
        train_texts, train_labels, train_domains,
        vocab=vocab, chunk_size=chunk_size, stride=stride,
    )
    val_ds = StressDataset(
        val_texts, val_labels, val_domains,
        vocab=vocab, chunk_size=chunk_size, stride=stride,
    )
    test_ds = StressDataset(
        test_texts, test_labels, test_domains,
        vocab=vocab, chunk_size=chunk_size, stride=stride,
    )

    train_loader = DataLoader(
        train_ds, batch_size=batch_size, shuffle=True, collate_fn=collate_fn,
    )
    val_loader = DataLoader(
        val_ds, batch_size=batch_size, shuffle=False, collate_fn=collate_fn,
    )
    test_loader = DataLoader(
        test_ds, batch_size=batch_size, shuffle=False, collate_fn=collate_fn,
    )

    return train_loader, val_loader, test_loader, vocab