File size: 7,306 Bytes
fef33ce | 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 | """
Feature matrix builder.
Assembles the full user-item feature matrix for the classical (LightGBM) model
by joining user features, item features, and interaction-level features.
Feature tiers (for the ablation experiment):
id_only : user_idx, item_idx (encoded as floats)
id_metadata : + item stats, category, brand, price
id_metadata_text : + user-item title cosine similarity
id_metadata_text_history : + user history aggregates
The `feature_set` parameter controls which tier is used.
"""
from __future__ import annotations
import pickle
from pathlib import Path
import numpy as np
import pandas as pd
from sklearn.feature_extraction.text import TfidfVectorizer
from src.config import DataConfig, PROCESSED_DIR
from src.features.item_features import compute_item_features
from src.features.text_features import (
build_item_tfidf,
build_user_text_profiles,
compute_text_similarity,
)
from src.features.user_features import compute_user_features
from src.utils.logging import get_logger
logger = get_logger(__name__)
FEATURE_SETS = ["id_only", "id_metadata", "id_metadata_text", "id_metadata_text_history"]
def build_feature_matrix(
df: pd.DataFrame,
user_features: pd.DataFrame,
item_features: pd.DataFrame,
user_profiles: pd.DataFrame,
vectorizer: TfidfVectorizer,
feature_set: str = "id_metadata_text_history",
) -> tuple[pd.DataFrame, list[str]]:
"""
Build a feature matrix for a given interaction DataFrame.
Args:
df: Interactions DataFrame with [user_idx, item_idx, label].
user_features: Per-user features (output of compute_user_features).
item_features: Per-item features (output of compute_item_features).
user_profiles: User text profiles (output of build_user_text_profiles).
vectorizer: Fitted TF-IDF vectorizer.
feature_set: One of FEATURE_SETS.
Returns:
Tuple of (feature_df, feature_column_names).
"""
if feature_set not in FEATURE_SETS:
raise ValueError(f"feature_set must be one of {FEATURE_SETS}, got {feature_set!r}")
feat = df[["user_idx", "item_idx"]].copy()
feature_cols: list[str] = []
# --- Tier 1: ID-only ---
feat["user_idx_f"] = feat["user_idx"].astype(float)
feat["item_idx_f"] = feat["item_idx"].astype(float)
feature_cols += ["user_idx_f", "item_idx_f"]
if feature_set == "id_only":
feat["label"] = df["label"].values
return feat[feature_cols + ["label"]], feature_cols
# --- Tier 2: + metadata ---
item_meta_cols = [
"item_avg_rating", "item_rating_std", "item_num_ratings",
"item_popularity", "item_positive_rate",
"category_idx", "brand_idx", "price_norm",
]
item_meta_cols = [c for c in item_meta_cols if c in item_features.columns]
feat = feat.merge(
item_features[["item_idx"] + item_meta_cols],
on="item_idx", how="left",
)
feature_cols += item_meta_cols
# Category match: does item's category == user's top category?
if "user_top_category" in user_features.columns and "category" in item_features.columns:
uid_to_top_cat = dict(zip(user_features["user_idx"], user_features["user_top_category"]))
iid_to_cat = dict(zip(item_features["item_idx"], item_features["category"]))
feat["category_match"] = [
1 if uid_to_top_cat.get(u, "") == iid_to_cat.get(i, "__") else 0
for u, i in zip(feat["user_idx"], feat["item_idx"])
]
feature_cols.append("category_match")
if feature_set == "id_metadata":
feat["label"] = df["label"].values
return feat[feature_cols + ["label"]], feature_cols
# --- Tier 3: + text similarity ---
text_sims = compute_text_similarity(user_profiles, item_features, vectorizer, feat)
feat["text_similarity"] = text_sims.values
feature_cols.append("text_similarity")
if feature_set == "id_metadata_text":
feat["label"] = df["label"].values
return feat[feature_cols + ["label"]], feature_cols
# --- Tier 4: + user history aggregates ---
user_hist_cols = [
"user_avg_rating", "user_rating_std", "user_num_ratings",
"user_num_positives", "user_n_categories", "user_recency",
"user_avg_item_price",
]
user_hist_cols = [c for c in user_hist_cols if c in user_features.columns]
feat = feat.merge(
user_features[["user_idx"] + user_hist_cols],
on="user_idx", how="left",
)
feature_cols += user_hist_cols
feat["label"] = df["label"].values
feat = feat.fillna(0.0)
return feat[feature_cols + ["label"]], feature_cols
def run_feature_pipeline(cfg: DataConfig) -> None:
"""
Build and save feature matrices for train / val / test splits.
Also saves user features, item features, and the TF-IDF vectorizer.
Args:
cfg: DataConfig.
"""
logger.info("Loading processed data…")
train_df = pd.read_parquet(cfg.train_path)
val_df = pd.read_parquet(cfg.val_path)
test_df = pd.read_parquet(cfg.test_path)
metadata_df = pd.read_parquet(cfg.metadata_path)
# Ensure label column exists in val/test
for split in (val_df, test_df):
if "label" not in split.columns:
split["label"] = (split["rating"] >= cfg.positive_rating_threshold).astype(int)
logger.info("Computing user features…")
user_features = compute_user_features(train_df, metadata_df, cfg.positive_rating_threshold)
logger.info("Computing item features…")
item_features, cat_enc, brand_enc = compute_item_features(
train_df, metadata_df, cfg.positive_rating_threshold
)
logger.info("Building TF-IDF and user text profiles…")
tfidf_matrix, vectorizer = build_item_tfidf(item_features)
user_profiles = build_user_text_profiles(train_df, item_features, cfg.positive_rating_threshold)
logger.info("Building feature matrices (full tier)…")
train_feat, feature_cols = build_feature_matrix(
train_df, user_features, item_features, user_profiles, vectorizer,
feature_set="id_metadata_text_history",
)
val_feat, _ = build_feature_matrix(
val_df, user_features, item_features, user_profiles, vectorizer,
feature_set="id_metadata_text_history",
)
test_feat, _ = build_feature_matrix(
test_df, user_features, item_features, user_profiles, vectorizer,
feature_set="id_metadata_text_history",
)
# Save
PROCESSED_DIR.mkdir(parents=True, exist_ok=True)
user_features.to_parquet(cfg.user_features_path, index=False)
item_features.to_parquet(cfg.item_features_path, index=False)
train_feat.to_parquet(cfg.train_features_path, index=False)
val_feat.to_parquet(cfg.val_features_path, index=False)
test_feat.to_parquet(cfg.test_features_path, index=False)
artifacts_path = PROCESSED_DIR / "feature_artifacts.pkl"
with open(artifacts_path, "wb") as f:
pickle.dump({
"vectorizer": vectorizer,
"cat_enc": cat_enc,
"brand_enc": brand_enc,
"feature_cols": feature_cols,
}, f)
logger.info("Feature pipeline complete. Saved to %s", PROCESSED_DIR)
logger.info("Feature columns (%d): %s", len(feature_cols), feature_cols)
|