residual autocorrelation

residual autocorrelation

# Compute Residual
train_pred = stepwise_model.predict(n_periods=106)
r_train = train - train_pred
r_test = test - pred
residual = pd.DataFrame(np.concatenate((r_train,r_test)), columns={"y"})


# Generate lag of Residuals from 1 step to 52 steps
# Adding the lag of the target variable from 1 steps back up to 52 
for i in range(1, 53):
    residual["lag_{}".format(i)] = residual.y.shift(i)

# Compute correlation of the Residual series and its lags
lag_corr = residual.corr()
lag_corr = lag_corr.iloc[1:,0]
lag_corr.columns = ["corr"]
order = lag_corr.abs().sort_values(ascending = False)
lag_corr = lag_corr[order.index]

# Plot the Residual Autocorrelation
plt.figure(figsize=(12, 6))
lag_corr.plot(kind='bar')
plt.grid(True, axis='y')
plt.title("Autocorrelation")
plt.hlines(y=0, xmin=0, xmax=len(lag_corr), linestyles='dashed')

# Plot other Criteria (Distribution, Variance, Residual mean)
# Residual mean and Distribution
print("Residual mean: ",residual.iloc[:,0].mean())
plt.hist(residual.iloc[:,0], bins=20)
plt.title("Residual Distribution")

# Residual variance plt.plot(residual.iloc[:,0]) 
plt.title("Residual") 
plt.hlines(y=0, xmin=0, xmax=len(residual), linestyles='dashed')