Feature-Importance und Gradient Boosting

Maschinelles Lernen für Finanzen in Python

Nathan George

Data Science Professor

Wochentag-Split

200‑Tage-SMA-Split

Feature-Importances extrahieren

from sklearn.ensemble import RandomForestRegressor

random_forest = RandomForestRegressor()
random_forest.fit(train_features, train_targets)

feature_importances = random_forest.feature_importances_

print(feature_importances)

[0.07586547 0.10697602 0.12215955 0.23969227 0.29010304 0.0314028
 0.11977058 0.00276721 0.00246329 0.0026431  0.00615667]

Sortieren und plotten

# feature importances from random forest model
importances = random_forest.feature_importances_

# index of greatest to least feature importances
sorted_index = np.argsort(importances)[::-1]

x = range(len(importances))
# create tick labels
labels = np.array(feature_names)[sorted_index]

plt.bar(x, importances[sorted_index], tick_label=labels)

# rotate tick labels to vertical
plt.xticks(rotation=90)
plt.show()

Feature-Importance-Plot

Lineare Modelle vs. Gradient Boosting

Camry vs. Helicopter

https://blog.kaggle.com/2017/01/23/a-kaggle-master-explains-gradient-boosting/

Boosting-Diagramm

Boosted Models

Verfügbare Boosted Models:

Gradient Boosting
Adaboost

Ein Gradient-Boosting-Modell fitten

from sklearn.ensemble import GradientBoostingRegressor

gbr = GradientBoostingRegressor(max_features=4,
                                learning_rate=0.01,
                                n_estimators=200,
                                subsample=0.6,
                                random_state=42)

gbr.fit(train_features, train_targets)

Get boosted!

Maschinelles Lernen für Finanzen in Python