[PYTHON] xgboost: Effektives Modell für maschinelles Lernen für Tabellendaten

xgboost: Hochleistungsklassifizierungs- / Vorhersagemodell durch Gradientenverstärkung des Baums. Sehr beliebt bei Kaggle.

Referenz

• Überblick über die Theorie yh0sh
• Kommentar Blog zaburo
• [Richtig verwenden mit Deep Learning @quora](https://www.quora.com/Why-is-xgboost-given-so-much-less-attention-than-deep-learning-despite-its-ubiquity-in -gewinnende-Kaggle-Lösungen)

Installation (Offizielles Dokument)

@mac

$ cd <workspace>
$ git clone --recursive https://github.com/dmlc/xgboost
$ cd xgboost; cp make/minimum.mk ./config.mk; make -j4
$ cd python-package; sudo python setup.py install

@ubuntu

$ cd <workspace>
$ git clone --recursive https://github.com/dmlc/xgboost
$ cd xgboost; make -j4
$ cd python-package; sudo python setup.py install

Verwendung 1: Regressionsmodell

regressor.py

import xgboost as xgb
from sklearn.model_selection import GridSearchCV
from sklearn.datasets import load_boston
from sklearn.metrics import mean_squared_error


#Daten gelesen
boston = load_boston()
X_train, X_test = boston.data[:400], boston.data[400:]
y_train, y_test = boston.target[:400], boston.target[400:]

#Erstellen eines xgboost-Modells
reg = xgb.XGBRegressor()

#Hyperparametersuche
reg_cv = GridSearchCV(reg, {'max_depth': [2,4,6], 'n_estimators': [50,100,200]}, verbose=1)
reg_cv.fit(X_train, y_train)
print reg_cv.best_params_, reg_cv.best_score_

#Lernen Sie erneut mit optimalen Parametern
reg = xgb.XGBRegressor(**reg_cv.best_params_)
reg.fit(X_train, y_train)

#Trainingsmodell speichern und laden
# import pickle
# pickle.dump(reg, open("model.pkl", "wb"))
# reg = pickle.load(open("model.pkl", "rb"))

#Bewertung des Lernmodells
pred_train = reg.predict(X_train)
pred_test = reg.predict(X_test)
print mean_squared_error(y_train, pred_train)
print mean_squared_error(y_test, pred_test)

#Plot von Feature-Wichtigkeit
import pandas as pd
import matplotlib.pyplot as plt
importances = pd.Series(reg.feature_importances_, index = boston.feature_names)
importances = importances.sort_values()
importances.plot(kind = "barh")
plt.title("imporance in the xgboost Model")
plt.show()

Verwendung 2: Klassifizierungsmodell

classifier.py

import xgboost as xgb
from sklearn.model_selection import GridSearchCV
from sklearn.datasets import load_digits
from sklearn.metrics import confusion_matrix, classification_report

#Daten gelesen
digits = load_digits()
X_train, X_test = digits.data[:1000], digits.data[1000:]
y_train, y_test = digits.target[:1000], digits.target[1000:]

#Erstellen eines xgboost-Modells
clf = xgb.XGBClassifier()

#Hyperparametersuche
clf_cv = GridSearchCV(clf, {'max_depth': [2,4,6], 'n_estimators': [50,100,200]}, verbose=1)
clf_cv.fit(X_train, y_train)
print clf_cv.best_params_, clf_cv.best_score_

#Lernen Sie erneut mit optimalen Parametern
clf = xgb.XGBClassifier(**clf_cv.best_params_)
clf.fit(X_train, y_train)

#Trainingsmodell speichern und laden
# import pickle
# pickle.dump(clf, open("model.pkl", "wb"))
# clf = pickle.load(open("model.pkl", "rb"))

#Bewertung des Lernmodells
pred = clf.predict(X_test)
print confusion_matrix(y_test, pred)
print classification_report(y_test, pred)

#              precision    recall  f1-score   support
# 
#           0       0.94      0.97      0.96        79
#           1       0.90      0.79      0.84        80
#           2       0.99      0.88      0.93        77
#           3       0.89      0.82      0.86        79
#           4       0.94      0.90      0.92        83
#           5       0.92      0.95      0.93        82
#           6       0.95      0.97      0.96        80
#           7       0.96      0.96      0.96        80
#           8       0.82      0.91      0.86        76
#           9       0.79      0.90      0.84        81
# 
# avg / total       0.91      0.91      0.91       797