Lære Gradient Boosting for Time Series | Classical ML Models for Time Series

Gradient boosting is an ensemble learning technique that builds models sequentially, with each new model correcting the errors of the previous ones. Unlike a single decision tree, which can be prone to overfitting or underfitting, gradient boosting combines many weak learners—typically shallow trees—into a strong predictive model. This approach is especially powerful for time series forecasting, where capturing subtle temporal patterns and non-linear relationships is crucial. Gradient boosting methods, such as scikit-learn's HistGradientBoostingRegressor, are robust to outliers and can handle complex feature interactions, making them well-suited for forecasting tasks when compared to single trees.


              12345678910111213141516171819202122232425262728293031323334353637383940414243444546
            
import numpy as np
import pandas as pd
from sklearn.ensemble import HistGradientBoostingRegressor
from sklearn.model_selection import TimeSeriesSplit
from sklearn.metrics import mean_squared_error
import matplotlib.pyplot as plt

# Create a synthetic time series (sinusoidal with noise)
np.random.seed(42)
n = 300
t = np.arange(n)
y = np.sin(0.04 * t) + np.random.normal(scale=0.3, size=n)
df = pd.DataFrame({'y': y})

# Create lagged features and rolling means
df['lag1'] = df['y'].shift(1)
df['lag2'] = df['y'].shift(2)
df['roll3'] = df['y'].rolling(window=3).mean().shift(1)
df['roll7'] = df['y'].rolling(window=7).mean().shift(1)
df = df.dropna()

X = df[['lag1', 'lag2', 'roll3', 'roll7']]
y = df['y']

tscv = TimeSeriesSplit(n_splits=5)
mse_scores = []

# For plotting last fold
last_train_idx, last_test_idx = None, None
last_predictions = None

for train_idx, test_idx in tscv.split(X):
    X_train, X_test = X.iloc[train_idx], X.iloc[test_idx]
    y_train, y_test = y.iloc[train_idx], y.iloc[test_idx]
    
    model = HistGradientBoostingRegressor(max_iter=100)
    model.fit(X_train, y_train)
    
    preds = model.predict(X_test)
    mse = mean_squared_error(y_test, preds)
    mse_scores.append(mse)
    
    last_train_idx, last_test_idx = train_idx, test_idx
    last_predictions = preds

print("Mean MSE across folds:", np.mean(mse_scores))


              1234567891011121314151617181920
            
# Visualize
plt.figure(figsize=(14, 6))

# Actual values (full series)
plt.plot(y.index, y.values, label="Actual", color="black")

# Predictions on final fold
plt.plot(y.index[last_test_idx], last_predictions, 
         label="Predictions (Last Fold)", color="orange", linewidth=2)

# Train-test split line
plt.axvline(y.index[last_test_idx][0], linestyle="--", color="gray", label="Train/Test Split")

plt.title("HistGradientBoosting TS Forecast - Last Fold Visualization")
plt.xlabel("Time")
plt.ylabel("Value")
plt.legend()
plt.grid(True)
plt.tight_layout()
plt.show()

When comparing random forests and gradient boosting for time series forecasting, several differences stand out:

Random forests average predictions from many independent trees;
This averaging reduces variance and improves robustness;
Some bias may remain if the underlying patterns are complex.

Gradient boosting takes a different approach:

Trees are built sequentially, with each tree correcting the errors of the previous one;
This process often leads to lower bias and better performance on challenging forecasting tasks;
Boosting models can be more sensitive to noise and overfitting if not tuned carefully.

Interpretability also differs:

Random forests are typically easier to analyze, as feature importances are more stable and the ensemble is less sensitive to small data changes;
Boosting models, while potentially more accurate, may require more careful examination to understand their predictions and feature dependencies.

1. What is a key advantage of gradient boosting over random forests for time series forecasting?

2. Which features are typically most important for boosting models in time series?

Var alt klart?

Tak for dine kommentarer!

Sektion 2. Kapitel 2

Spørg AI

Spørg om hvad som helst eller prøv et af de foreslåede spørgsmål for at starte vores chat

Suggested prompts:

Can you explain how the lagged features and rolling means help with time series forecasting?

What are some best practices for tuning HistGradientBoostingRegressor for time series data?

How does the performance of gradient boosting compare to random forests in this example?

Stryg for at vise menuen


              12345678910111213141516171819202122232425262728293031323334353637383940414243444546
            
import numpy as np
import pandas as pd
from sklearn.ensemble import HistGradientBoostingRegressor
from sklearn.model_selection import TimeSeriesSplit
from sklearn.metrics import mean_squared_error
import matplotlib.pyplot as plt

# Create a synthetic time series (sinusoidal with noise)
np.random.seed(42)
n = 300
t = np.arange(n)
y = np.sin(0.04 * t) + np.random.normal(scale=0.3, size=n)
df = pd.DataFrame({'y': y})

# Create lagged features and rolling means
df['lag1'] = df['y'].shift(1)
df['lag2'] = df['y'].shift(2)
df['roll3'] = df['y'].rolling(window=3).mean().shift(1)
df['roll7'] = df['y'].rolling(window=7).mean().shift(1)
df = df.dropna()

X = df[['lag1', 'lag2', 'roll3', 'roll7']]
y = df['y']

tscv = TimeSeriesSplit(n_splits=5)
mse_scores = []

# For plotting last fold
last_train_idx, last_test_idx = None, None
last_predictions = None

for train_idx, test_idx in tscv.split(X):
    X_train, X_test = X.iloc[train_idx], X.iloc[test_idx]
    y_train, y_test = y.iloc[train_idx], y.iloc[test_idx]
    
    model = HistGradientBoostingRegressor(max_iter=100)
    model.fit(X_train, y_train)
    
    preds = model.predict(X_test)
    mse = mean_squared_error(y_test, preds)
    mse_scores.append(mse)
    
    last_train_idx, last_test_idx = train_idx, test_idx
    last_predictions = preds

print("Mean MSE across folds:", np.mean(mse_scores))


              1234567891011121314151617181920
            
# Visualize
plt.figure(figsize=(14, 6))

# Actual values (full series)
plt.plot(y.index, y.values, label="Actual", color="black")

# Predictions on final fold
plt.plot(y.index[last_test_idx], last_predictions, 
         label="Predictions (Last Fold)", color="orange", linewidth=2)

# Train-test split line
plt.axvline(y.index[last_test_idx][0], linestyle="--", color="gray", label="Train/Test Split")

plt.title("HistGradientBoosting TS Forecast - Last Fold Visualization")
plt.xlabel("Time")
plt.ylabel("Value")
plt.legend()
plt.grid(True)
plt.tight_layout()
plt.show()