File size: 8,685 Bytes
e323466
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
"""Dataset loading, cleaning, and splitting for customer support intent classification."""

import os
import sys
from pathlib import Path
from typing import Dict, Tuple

import pandas as pd
from loguru import logger
from sklearn.model_selection import train_test_split

from src.data.preprocessing import clean_texts, set_global_seeds

INTENT_CATEGORIES = [
    "billing_issue",
    "account_access",
    "technical_support",
    "product_inquiry",
    "cancellation_request",
    "general_feedback",
]

# Maps Bitext intent tags (lowercased) → our 6 categories.
# Bitext uses tags like "check_invoice", "cancel_order", etc.
LABEL_MAP: Dict[str, str] = {
    # billing_issue
    "check_invoice": "billing_issue",
    "payment_issue": "billing_issue",
    "check_payment_methods": "billing_issue",
    "check_refund_policy": "billing_issue",
    "get_refund": "billing_issue",
    "track_refund": "billing_issue",
    "check_cancellation_fee": "billing_issue",
    "registration_problems": "billing_issue",
    # account_access
    "change_password": "account_access",
    "recover_password": "account_access",
    "edit_account": "account_access",
    "delete_account": "account_access",
    "create_account": "account_access",
    "switch_account": "account_access",
    "set_up_shipping_address": "account_access",
    # technical_support
    "complaint": "technical_support",
    "delivery_options": "technical_support",
    "delivery_period": "technical_support",
    "track_order": "technical_support",
    "place_order": "technical_support",
    "change_order": "technical_support",
    "check_invoices": "billing_issue",
    # product_inquiry
    "get_invoice": "product_inquiry",
    "check_payment_methods": "product_inquiry",
    "newsletter_subscription": "product_inquiry",
    "product_compatibility": "product_inquiry",
    "review": "product_inquiry",
    "check_warranty_guarantee": "product_inquiry",
    # cancellation_request
    "cancel_order": "cancellation_request",
    "cancel_subscription": "cancellation_request",
    "return": "cancellation_request",
    "contact_human_agent": "cancellation_request",
    # general_feedback
    "contact_customer_service": "general_feedback",
    "check_contact_payment_methods": "general_feedback",
    "feedback": "general_feedback",
    "question": "general_feedback",
}


def _load_from_huggingface(dataset_name: str) -> pd.DataFrame:
    """Load dataset from Hugging Face hub and return as a combined DataFrame."""
    from datasets import load_dataset  # type: ignore

    logger.info(f"Loading dataset '{dataset_name}' from HuggingFace Hub…")
    ds = load_dataset(dataset_name, trust_remote_code=True)

    # Combine all splits into one DataFrame for re-splitting
    frames = []
    for split_name, split_data in ds.items():
        df = split_data.to_pandas()
        frames.append(df)
    df = pd.concat(frames, ignore_index=True)
    logger.info(f"Loaded {len(df):,} rows from HuggingFace. Columns: {list(df.columns)}")
    return df


def _map_labels(df: pd.DataFrame) -> pd.DataFrame:
    """Map raw Bitext intent labels to the 6 target categories."""
    # Detect the intent column name
    intent_col = None
    for col in ["intent", "label", "category", "tag"]:
        if col in df.columns:
            intent_col = col
            break
    if intent_col is None:
        raise ValueError(f"Could not find intent column. Available: {list(df.columns)}")

    logger.info(f"Using '{intent_col}' as the intent column.")
    raw_labels = df[intent_col].str.lower().str.strip()

    # Direct mapping
    mapped = raw_labels.map(LABEL_MAP)

    # For unmapped labels try substring matching
    unmapped_mask = mapped.isna()
    if unmapped_mask.any():
        def _fallback(raw: str) -> str:
            for keyword, category in [
                ("bill", "billing_issue"),
                ("payment", "billing_issue"),
                ("refund", "billing_issue"),
                ("invoice", "billing_issue"),
                ("password", "account_access"),
                ("account", "account_access"),
                ("login", "account_access"),
                ("technical", "technical_support"),
                ("delivery", "technical_support"),
                ("track", "technical_support"),
                ("order", "technical_support"),
                ("product", "product_inquiry"),
                ("item", "product_inquiry"),
                ("warranty", "product_inquiry"),
                ("cancel", "cancellation_request"),
                ("return", "cancellation_request"),
                ("feedback", "general_feedback"),
                ("complaint", "general_feedback"),
            ]:
                if keyword in raw:
                    return category
            return "general_feedback"

        mapped[unmapped_mask] = raw_labels[unmapped_mask].apply(_fallback)
        logger.warning(
            f"Fallback mapping applied to {unmapped_mask.sum()} rows."
        )

    df = df.copy()
    # Detect text column
    text_col = None
    for col in ["utterance", "text", "instruction", "input", "query", "sentence"]:
        if col in df.columns:
            text_col = col
            break
    if text_col is None:
        raise ValueError(f"Could not find text column. Available: {list(df.columns)}")

    logger.info(f"Using '{text_col}' as the text column.")
    df["text"] = df[text_col].astype(str)
    df["label"] = mapped.astype(str)
    return df[["text", "label"]]


def load_and_prepare(
    dataset_name: str,
    processed_dir: str,
    train_ratio: float = 0.70,
    val_ratio: float = 0.15,
    seed: int = 42,
) -> Tuple[pd.DataFrame, pd.DataFrame, pd.DataFrame]:
    """Full pipeline: load → clean → map labels → split → save.

    Args:
        dataset_name: HuggingFace dataset identifier.
        processed_dir: Directory to save CSV splits.
        train_ratio: Fraction of data for training.
        val_ratio: Fraction of data for validation.
        seed: Random seed.

    Returns:
        Tuple of (train_df, val_df, test_df).
    """
    set_global_seeds(seed)
    Path(processed_dir).mkdir(parents=True, exist_ok=True)

    df = _load_from_huggingface(dataset_name)
    df = _map_labels(df)

    # Clean text
    logger.info("Cleaning text…")
    df["text"] = clean_texts(df["text"].tolist())
    df = df[df["text"].str.len() > 0].reset_index(drop=True)

    # Check class counts
    counts = df["label"].value_counts()
    logger.info(f"Label distribution before split:\n{counts.to_string()}")
    for cat in INTENT_CATEGORIES:
        n = counts.get(cat, 0)
        if n < 50:
            logger.warning(f"Category '{cat}' has only {n} examples (< 50).")

    # Stratified train/val/test split
    test_ratio = 1.0 - train_ratio - val_ratio
    train_df, temp_df = train_test_split(
        df,
        test_size=(val_ratio + test_ratio),
        stratify=df["label"],
        random_state=seed,
    )
    relative_val = val_ratio / (val_ratio + test_ratio)
    val_df, test_df = train_test_split(
        temp_df,
        test_size=(1.0 - relative_val),
        stratify=temp_df["label"],
        random_state=seed,
    )

    for name, split in [("train", train_df), ("val", val_df), ("test", test_df)]:
        path = Path(processed_dir) / f"{name}.csv"
        split.to_csv(path, index=False)
        logger.info(f"Saved {name} ({len(split):,} rows) → {path}")
        dist = split["label"].value_counts()
        logger.info(f"  {name} distribution:\n{dist.to_string()}")

    return train_df, val_df, test_df


def load_splits(processed_dir: str) -> Tuple[pd.DataFrame, pd.DataFrame, pd.DataFrame]:
    """Load pre-saved CSV splits.

    Args:
        processed_dir: Directory containing train.csv, val.csv, test.csv.

    Returns:
        Tuple of (train_df, val_df, test_df).
    """
    base = Path(processed_dir)
    train_df = pd.read_csv(base / "train.csv")
    val_df = pd.read_csv(base / "val.csv")
    test_df = pd.read_csv(base / "test.csv")
    logger.info(
        f"Loaded splits — train: {len(train_df):,}, val: {len(val_df):,}, test: {len(test_df):,}"
    )
    return train_df, val_df, test_df


if __name__ == "__main__":
    import yaml

    logging_path = Path("logs")
    logging_path.mkdir(exist_ok=True)
    logger.add(logging_path / "dataset.log", rotation="10 MB")

    with open("config/config.yaml") as f:
        cfg = yaml.safe_load(f)

    load_and_prepare(
        dataset_name=cfg["data"]["dataset_name"],
        processed_dir=cfg["paths"]["data_processed"],
        train_ratio=cfg["data"]["train_ratio"],
        val_ratio=cfg["data"]["val_ratio"],
        seed=cfg["data"]["seed"],
    )