residual_diagnostics.png

# --- Regression Residual Diagnostics ---
print("[12] Regressions-Residuendiagnostik ...")

np.random.seed(42)
fitted_vals = np.linspace(10, 90, 200)

# 1. Homoscedasticity (Good model): residuals are a constant band
residuals_good = np.random.normal(0, 5, len(fitted_vals))

# 2. Heteroscedasticity (Violated model): variance increases with fitted values (funnel shape)
residuals_bad = np.random.normal(0, 1.0 + 0.25 * fitted_vals, len(fitted_vals))

fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(10, 4.5), sharey=True)

# Homoscedasticity
ax1.scatter(fitted_vals, residuals_good, color=PRIMARY, alpha=0.6, edgecolors="none")
ax1.axhline(0, color="#16181C", linestyle="--", linewidth=1.2)
ax1.set_xlabel("Vorhergesagte Werte (Fitted Values)")
ax1.set_ylabel("Residuen (Residuals)")
ax1.set_title("Homoskedastizität (Annahme erfüllt)\n(Gleichmäßige Streuung)")
ax1.grid(True, linestyle=":", alpha=0.6)

# Heteroscedasticity
ax2.scatter(fitted_vals, residuals_bad, color=EVENT, alpha=0.6, edgecolors="none")
ax2.axhline(0, color="#16181C", linestyle="--", linewidth=1.2)
ax2.set_xlabel("Vorhergesagte Werte (Fitted Values)")
ax2.set_title("Heteroskedastizität (Annahme verletzt)\n(Trichterförmige Streuung)")
ax2.grid(True, linestyle=":", alpha=0.6)

plt.tight_layout()
(ASSETS / "12-regression" / "assets").mkdir(parents=True, exist_ok=True)
save(fig, ASSETS / "12-regression" / "assets" / "residual_diagnostics.png")