File size: 3,849 Bytes
fef33ce 09fcfe1 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 | """
User-level feature engineering.
Computes aggregate statistics over a user's interaction history:
- avg_rating, rating_std
- num_ratings (activity level)
- avg_review_recency (days since first interaction, normalised)
- top_categories (mode category in rated items)
- category_affinity vector (fraction of positive ratings per category)
All features are derived from training data only to prevent leakage.
"""
from __future__ import annotations
import numpy as np
import pandas as pd
from src.utils.logging import get_logger
logger = get_logger(__name__)
def compute_user_features(
train_df: pd.DataFrame,
metadata_df: pd.DataFrame,
positive_threshold: float = 4.0,
) -> pd.DataFrame:
"""
Compute per-user aggregate features from the training split.
Args:
train_df: Training interactions with columns [user_idx, item_idx, rating, timestamp].
metadata_df: Item metadata with columns [item_idx, category, brand, price].
positive_threshold: Rating threshold for a positive interaction.
Returns:
DataFrame indexed by user_idx with one row per user.
"""
# Merge metadata into interactions
merged = train_df.merge(
metadata_df[["item_idx", "category", "brand", "price"]],
on="item_idx",
how="left",
)
merged["is_positive"] = (merged["rating"] >= positive_threshold).astype(int)
# Basic rating stats
rating_stats = (
merged.groupby("user_idx")["rating"]
.agg(["mean", "std", "count"])
.rename(columns={"mean": "user_avg_rating", "std": "user_rating_std", "count": "user_num_ratings"})
)
rating_stats["user_rating_std"] = rating_stats["user_rating_std"].fillna(0.0)
# Positive interaction count
pos_count = (
merged[merged["is_positive"] == 1]
.groupby("user_idx")["item_idx"]
.count()
.rename("user_num_positives")
)
# Most frequent category among positively-rated items
pos_merged = merged[merged["is_positive"] == 1].copy()
top_category = (
pos_merged.groupby("user_idx")["category"]
.agg(lambda x: x.mode().iat[0] if len(x) > 0 and len(x.mode()) > 0 else "Unknown")
.rename("user_top_category")
)
# Number of unique categories interacted with
n_categories = (
merged.groupby("user_idx")["category"]
.nunique()
.rename("user_n_categories")
)
# Recency: mean timestamp normalised to [0, 1]
if merged["timestamp"].notna().any():
ts_min = merged["timestamp"].min()
ts_max = merged["timestamp"].max()
ts_range = max(ts_max - ts_min, 1)
recency = (
merged.groupby("user_idx")["timestamp"]
.mean()
.apply(lambda t: (t - ts_min) / ts_range)
.rename("user_recency")
)
else:
recency = pd.Series(0.5, index=rating_stats.index, name="user_recency")
# Average price of positively rated items
avg_price = (
pos_merged.groupby("user_idx")["price"]
.mean()
.rename("user_avg_item_price")
)
user_features = (
rating_stats
.join(pos_count, how="left")
.join(top_category, how="left")
.join(n_categories, how="left")
.join(recency, how="left")
.join(avg_price, how="left")
)
user_features["user_num_positives"] = user_features["user_num_positives"].fillna(0)
user_features["user_top_category"] = user_features["user_top_category"].fillna("Unknown")
user_features["user_avg_item_price"] = user_features["user_avg_item_price"].fillna(
user_features["user_avg_item_price"].median()
)
logger.info("Computed user features for %d users, %d features",
len(user_features), len(user_features.columns))
return user_features.reset_index()
|