data_utils.py

import numpy as np

# -------------------- Normalization / Scaling --------------------

def minmax_scale(X, feature_range=(0.0, 1.0), eps=1e-12):
    """
    Scale each feature of X to [min, max] range (default [0,1]).
    Returns (X_scaled, X_min, X_max) so you can invert later.
    """
    X = np.asarray(X, dtype=float)
    X_min = X.min(axis=0)
    X_max = X.max(axis=0)
    scale = (feature_range[1] - feature_range[0]) / np.maximum(X_max - X_min, eps)
    X_scaled = feature_range[0] + (X - X_min) * scale
    return X_scaled, X_min, X_max


def minmax_inverse(X_scaled, X_min, X_max, feature_range=(0.0, 1.0), eps=1e-12):
    """
    Inverse of minmax_scale() — returns data in original feature scale.
    """
    scale = (X_max - X_min) / np.maximum(feature_range[1] - feature_range[0], eps)
    return X_min + (X_scaled - feature_range[0]) * scale


def zscore_normalize(X, eps=1e-12):
    """
    Normalize each feature to zero mean and unit variance (Z-score normalization).
    Returns (X_norm, mean, std).
    """
    X = np.asarray(X, dtype=float)
    mean = X.mean(axis=0)
    std = np.maximum(X.std(axis=0), eps)
    return (X - mean) / std, mean, std


def zscore_inverse(X_norm, mean, std, eps=1e-12):
    return mean + X_norm * np.maximum(std, eps)


def save_original_with_labels(instance_ids, X_raw, labels, out_csv, header):
    import csv
    import numpy as np

    if len(instance_ids) != len(X_raw) or len(labels) != len(X_raw):
        raise ValueError(
            f"Length mismatch: instance_ids={len(instance_ids)}, X_raw={len(X_raw)}, labels={len(labels)}"
        )

    with open(out_csv, "w", newline="") as f:
        w = csv.writer(f)
        w.writerow(header)

        for inst, row, lab in zip(instance_ids, X_raw, labels):
            x1 = float(row[0])                 # IPC
            x2 = int(np.rint(row[1]))          # TOT_INS -> force integer, remove .0000001 noise

            # Optional: round IPC for nicer CSV
            x1 = round(x1, 6)

            w.writerow([int(inst), x1, x2, int(lab)])