File size: 7,510 Bytes
66242b8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
from __future__ import annotations

from dataclasses import dataclass
from typing import Any, Optional

import numpy as np
import pandas as pd
from scipy import sparse
from sklearn.decomposition import TruncatedSVD


@dataclass(slots=True)
class Config:
    verbose: bool = True
    text_columns: tuple[str, ...] = ("text1", "text2")
    keep_original_columns: bool = False
    reduce_tfidf: bool = True
    reduce_char_ngrams: bool = True
    reduce_pos_ngrams: bool = True
    tfidf_components: int = 300
    char_components: int = 300
    pos_components: int = 100
    random_state: int = 42

config = Config()


# column suffixes for each features family/category
FAMILY_SPECS = {
    "tfidf": {
        "suffix_prefix": "tfidf_",
        "config_flag": "reduce_tfidf",
        "components_attr": "tfidf_components",
        "reduced_prefix": "tfidf_svd",
    },
    "char_ngrams": {
        "suffix_prefix": "char",
        "must_contain": "_tfidf_",
        "config_flag": "reduce_char_ngrams",
        "components_attr": "char_components",
        "reduced_prefix": "char_tfidf_svd",
    },
    "pos_ngrams": {
        "suffix_prefix": "pos",
        "must_contain": "_tfidf_",
        "config_flag": "reduce_pos_ngrams",
        "components_attr": "pos_components",
        "reduced_prefix": "pos_tfidf_svd",
    }}



def _match_family_suffix(suffix: str, family_name: str) -> bool:
    spec = FAMILY_SPECS[family_name]
    if family_name == "tfidf":
        return suffix.startswith(spec["suffix_prefix"])
    return suffix.startswith(spec["suffix_prefix"]) and spec["must_contain"] in suffix

# get features column suffixes
def discover_family_suffixes(df: pd.DataFrame, family_name: str) -> list[str]:
    suffixes: list[str] = []
    for column in df.columns:
        if not column.startswith("text1_"): # excluding original text columns
            continue
        suffix = column[len("text1_"):]
        if _match_family_suffix(suffix, family_name):
            suffixes.append(suffix)
    return suffixes


# ========================= Fit SVD ====================================

def _shared_train_matrix(train_df: pd.DataFrame, suffixes: list[str]) -> sparse.csr_matrix:
    text1_columns = [f"text1_{suffix}" for suffix in suffixes]
    text2_columns = [f"text2_{suffix}" for suffix in suffixes]
    train_text1 = sparse.csr_matrix(train_df[text1_columns].to_numpy(dtype=np.float32, copy=False))
    train_text2 = sparse.csr_matrix(train_df[text2_columns].to_numpy(dtype=np.float32, copy=False))
    return sparse.vstack([train_text1, train_text2], format="csr")

# truncatedSVD cannot use more components than the data matrix can support
# the safe upper bound is limited by the smaller matrix dimension.
def _effective_components(requested_components: int, train_matrix: sparse.csr_matrix) -> int:
    max_components = min(train_matrix.shape[0] - 1, train_matrix.shape[1] - 1)
    if max_components < 1: # guard against tiny matrices where upper bound would be 0
        return 1
    return min(requested_components, max_components)

def fit_family_svd(
    train_df: pd.DataFrame,
    family_name: str,
    suffixes: list[str],
    config: Config = config,
) -> tuple[TruncatedSVD, int]:
    train_matrix = _shared_train_matrix(train_df, suffixes) # building sparse matrix for fit
    requested_components = getattr(config, FAMILY_SPECS[family_name]["components_attr"]) # getting corresponding target components from config
    n_components = _effective_components(requested_components, train_matrix)

    svd = TruncatedSVD(n_components=n_components, random_state=config.random_state)
    svd.fit(train_matrix)
    return svd, n_components

# =================================================================


def transform_family_split(
    df: pd.DataFrame,
    family_name: str,
    suffixes: list[str],
    svd: TruncatedSVD,
    config: Config = config,
) -> pd.DataFrame:
    
    result = df.copy().reset_index(drop=True)
    text1_columns = [f"text1_{suffix}" for suffix in suffixes]
    text2_columns = [f"text2_{suffix}" for suffix in suffixes]

    text1_matrix = sparse.csr_matrix(result[text1_columns].to_numpy(dtype=np.float32, copy=False))
    text2_matrix = sparse.csr_matrix(result[text2_columns].to_numpy(dtype=np.float32, copy=False))

    # project the original high-dimensional features into the reduced latent space
    text1_reduced = svd.transform(text1_matrix)
    text2_reduced = svd.transform(text2_matrix)

    reduced_prefix = FAMILY_SPECS[family_name]["reduced_prefix"]
    text1_reduced_df = pd.DataFrame(text1_reduced,
        columns=[f"text1_{reduced_prefix}_{index:04d}" for index in range(text1_reduced.shape[1])])
    text2_reduced_df = pd.DataFrame(text2_reduced,
        columns=[f"text2_{reduced_prefix}_{index:04d}" for index in range(text2_reduced.shape[1])])

    if not config.keep_original_columns:
        result = result.drop(columns=text1_columns + text2_columns)

    result = pd.concat([
            result.reset_index(drop=True),
            text1_reduced_df.reset_index(drop=True),
            text2_reduced_df.reset_index(drop=True),
        ], axis=1)
    
    return result



def dimensionality_reduction_wrapper(
    dict_df: dict[str, pd.DataFrame],
    config: Config = config,
) -> tuple[dict[str, pd.DataFrame], pd.DataFrame, dict[str, Any]]:


    if config.verbose:
        print("======= DIMENSIONALITY REDUCTION START =======")

    reduced_dict_df = {split: df.copy().reset_index(drop=True) for split, df in dict_df.items()}
    summary_rows: list[dict[str, Any]] = []
    artifacts: dict[str, Any] = {"svd_models": {}, "family_suffixes": {}}

    for family_name, spec in FAMILY_SPECS.items():

        if not getattr(config, spec["config_flag"]):
            continue

        suffixes = discover_family_suffixes(reduced_dict_df["train"], family_name)
        if config.verbose:
            print(f"\nFitting TruncatedSVD for family='{family_name}' with {len(suffixes):,} base features")

        svd, n_components = fit_family_svd(reduced_dict_df["train"], family_name=family_name, suffixes=suffixes, config=config)

        artifacts["svd_models"][family_name] = svd
        artifacts["family_suffixes"][family_name] = suffixes

        for split, df in list(reduced_dict_df.items()):

            if config.verbose:
                print(f"    Transforming split='{split}' for family='{family_name}'")
            reduced_dict_df[split] = transform_family_split(
                df,
                family_name=family_name,
                suffixes=suffixes,
                svd=svd,
                config=config,
            )

        explained_variance = float(svd.explained_variance_ratio_.sum())
        summary_rows.append({
                "family": family_name,
                "original_base_features": len(suffixes),
                "reduced_components": n_components,
                "explained_variance_ratio_sum": round(explained_variance, 6),
            })

        if config.verbose:
            print(
                f"    Reduced family='{family_name}' to {n_components} components "
                f"(explained variance sum={explained_variance:.4f})"
            )

    reduction_summary_df = pd.DataFrame(summary_rows)

    if config.verbose:
        print("\nDimensionality reduction summary:")
        print(reduction_summary_df)
        print("")
        print("======= DIMENSIONALITY REDUCTION END =======")
        print("")

    return reduced_dict_df, reduction_summary_df, artifacts