preprocess.py

"""
Preprocessing for Customer Churn Predictor.
Converts a raw customer record (matching Telco-style schema) into the
one-hot encoded feature vector the trained model expects.
"""

import json
import pandas as pd

with open("schema.json") as f:
    SCHEMA = json.load(f)

MODEL_COLUMNS = SCHEMA["model_columns"]

BINARY_MAP = {"Yes": 1, "No": 0}
BINARY_COLS = ["Partner", "Dependents", "PhoneService", "PaperlessBilling"]
MULTI_CAT_COLS = [
    "gender", "MultipleLines", "InternetService", "OnlineSecurity",
    "OnlineBackup", "DeviceProtection", "TechSupport", "StreamingTV",
    "StreamingMovies", "Contract", "PaymentMethod",
]


def preprocess(record: dict) -> pd.DataFrame:
    """
    record: dict of raw customer fields (see schema.json -> raw_input_schema)
    returns: single-row DataFrame aligned to MODEL_COLUMNS, ready for model.predict_proba
    """
    df = pd.DataFrame([record])

    # TotalCharges: treat blank/missing as 0 (new customer, tenure 0 case)
    df["TotalCharges"] = pd.to_numeric(df.get("TotalCharges", 0), errors="coerce").fillna(0.0)

    for col in BINARY_COLS:
        df[col] = df[col].map(BINARY_MAP)

    df_encoded = pd.get_dummies(df, columns=MULTI_CAT_COLS)

    # Align to the exact training-time columns (missing dummy cols -> 0, drop extras)
    df_aligned = df_encoded.reindex(columns=MODEL_COLUMNS, fill_value=0)

    return df_aligned