features.py

import numpy as np
import pandas as pd

from model import model_columns
from schema import ShopperInput

MONTH_ORDER = {
    "Jan": 1, "Feb": 2, "Mar": 3, "Apr": 4,
    "May": 5, "June": 6, "Jul": 7, "Aug": 8,
    "Sep": 9, "Oct": 10, "Nov": 11, "Dec": 12,
}

LOG_COLUMNS = [
    "Administrative_Duration",
    "Informational_Duration",
    "ProductRelated_Duration",
    "TotalDuration",
    "ProductDurationPerPage",
    "AdminDurationPerPage",
    "InfoDurationPerPage",
    "AvgDurationPerPage",
    "ProductEngagement",
    "DurationWeightedExit",
    "DurationWeightedBounce",
]


def create_model_features(input_data: ShopperInput) -> pd.DataFrame:
    """Create the same engineered features that were used during training."""

    df = pd.DataFrame([input_data.model_dump()])

    base_numeric_columns = [
        "Administrative", "Administrative_Duration",
        "Informational", "Informational_Duration",
        "ProductRelated", "ProductRelated_Duration",
        "BounceRates", "ExitRates", "PageValues", "SpecialDay",
    ]
    for col in base_numeric_columns:
        df[col] = pd.to_numeric(df[col], errors="coerce")

    # Step 1: totals
    df["TotalDuration"] = (
        df["Administrative_Duration"].fillna(0)
        + df["Informational_Duration"].fillna(0)
        + df["ProductRelated_Duration"].fillna(0)
    )
    df["TotalPages"] = (
        df["Administrative"].fillna(0)
        + df["Informational"].fillna(0)
        + df["ProductRelated"].fillna(0)
    )

    # Step 2: averages and ratios
    df["ProductDurationPerPage"] = df["ProductRelated_Duration"] / df["ProductRelated"].replace(0, np.nan)
    df["AdminDurationPerPage"]   = df["Administrative_Duration"] / df["Administrative"].replace(0, np.nan)
    df["InfoDurationPerPage"]    = df["Informational_Duration"]  / df["Informational"].replace(0, np.nan)
    df["ProductPageRatio"]       = df["ProductRelated"]          / df["TotalPages"].replace(0, np.nan)
    df["AdminPageRatio"]         = df["Administrative"]          / df["TotalPages"].replace(0, np.nan)
    df["InfoPageRatio"]          = df["Informational"]           / df["TotalPages"].replace(0, np.nan)
    df["ProductDurationRatio"]   = df["ProductRelated_Duration"] / df["TotalDuration"].replace(0, np.nan)
    df["AdminDurationRatio"]     = df["Administrative_Duration"] / df["TotalDuration"].replace(0, np.nan)
    df["InfoDurationRatio"]      = df["Informational_Duration"]  / df["TotalDuration"].replace(0, np.nan)

    # Step 3: engagement signals
    df["AvgDurationPerPage"]     = df["TotalDuration"]   / df["TotalPages"].replace(0, np.nan)
    df["ExitBounceGap"]          = df["ExitRates"]       - df["BounceRates"]
    df["ExitBounceRatio"]        = df["ExitRates"]       / df["BounceRates"].replace(0, np.nan)
    df["ProductEngagement"]      = df["ProductRelated"]  * df["ProductDurationPerPage"]
    df["DurationWeightedExit"]   = df["TotalDuration"]   * (1 - df["ExitRates"])
    df["DurationWeightedBounce"] = df["TotalDuration"]   * (1 - df["BounceRates"])

    # Step 4: month features
    df["MonthNum"]        = df["Month"].map(MONTH_ORDER)
    df["IsHolidaySeason"] = df["Month"].isin(["Nov", "Dec"]).astype(int)

    # Step 5: log-transform skewed duration features
    for col in LOG_COLUMNS:
        df[col + "_log"] = np.log1p(df[col].clip(lower=0))

    return df[model_columns]