在 Kaggle 的不少比賽中,咱們能夠看到不少 winner 喜歡用 xgboost,並且得到很是好的表現,今天就來看看 xgboost 究竟是什麼以及如何應用。html
XGBoost :eXtreme Gradient Boosting
項目地址:https://github.com/dmlc/xgboostpython
是由 Tianqi Chen http://homes.cs.washington.edu/~tqchen/ 最初開發的實現可擴展,便攜,分佈式 gradient boosting (GBDT, GBRT or GBM) 算法的一個庫,能夠下載安裝並應用於 C++,Python,R,Julia,Java,Scala,Hadoop,如今有不少協做者共同開發維護。ios
XGBoost 所應用的算法就是 gradient boosting decision tree,既能夠用於分類也能夠用於迴歸問題中。git
那什麼是 Gradient Boosting?github
Gradient boosting 是 boosting 的其中一種方法web
所謂 Boosting ,就是將弱分離器 f_i(x) 組合起來造成強分類器 F(x) 的一種方法。算法
因此 Boosting 有三個要素:api
A loss function to be optimized:
例如分類問題中用 cross entropy,迴歸問題用 mean squared error。數組
A weak learner to make predictions:
例如決策樹。dom
An additive model:
將多個弱學習器累加起來組成強學習器,進而使目標損失函數達到極小。
Gradient boosting 就是經過加入新的弱學習器,來努力糾正前面全部弱學習器的殘差,最終這樣多個學習器相加在一塊兒用來進行最終預測,準確率就會比單獨的一個要高。之因此稱爲 Gradient,是由於在添加新模型時使用了梯度降低算法來最小化的損失。
前面已經知道,XGBoost 就是對 gradient boosting decision tree 的實現,可是通常來講,gradient boosting 的實現是比較慢的,由於每次都要先構造出一個樹並添加到整個模型序列中。
而 XGBoost 的特色就是計算速度快,模型表現好,這兩點也正是這個項目的目標。
表現快是由於它具備這樣的設計:
下圖就是 XGBoost 與其它 gradient boosting 和 bagged decision trees 實現的效果比較,能夠看出它比 R, Python,Spark,H2O 中的基準配置要更快。
另一個優勢就是在預測問題中模型表現很是好,下面是幾個 kaggle winner 的賽後採訪連接,能夠看出 XGBoost 的在實戰中的效果。
先來用 Xgboost 作一個簡單的二分類問題,如下面這個數據爲例,來判斷病人是否會在 5 年內患糖尿病,這個數據前 8 列是變量,最後一列是預測值爲 0 或 1。
數據描述:
https://archive.ics.uci.edu/ml/datasets/Pima+Indians+Diabetes
下載數據集,並保存爲 「pima-indians-diabetes.csv「 文件:
https://archive.ics.uci.edu/ml/machine-learning-databases/pima-indians-diabetes/pima-indians-diabetes.data
引入 xgboost 等包
from numpy import loadtxt from xgboost import XGBClassifier from sklearn.model_selection import train_test_split from sklearn.metrics import accuracy_score
分出變量和標籤
dataset = loadtxt('pima-indians-diabetes.csv', delimiter=",") X = dataset[:,0:8] Y = dataset[:,8]
將數據分爲訓練集和測試集,測試集用來預測,訓練集用來學習模型
seed = 7 test_size = 0.33 X_train, X_test, y_train, y_test = train_test_split(X, Y, test_size=test_size, random_state=seed)
xgboost 有封裝好的分類器和迴歸器,能夠直接用 XGBClassifier 創建模型
這裏是 XGBClassifier 的文檔:
http://xgboost.readthedocs.io/en/latest/python/python_api.html#module-xgboost.sklearn
model = XGBClassifier()
model.fit(X_train, y_train)
xgboost 的結果是每一個樣本屬於第一類的機率,須要用 round 將其轉換爲 0 1 值
y_pred = model.predict(X_test) predictions = [round(value) for value in y_pred]
獲得 Accuracy: 77.95%
accuracy = accuracy_score(y_test, predictions) print("Accuracy: %.2f%%" % (accuracy * 100.0))
xgboost 能夠在模型訓練時,評價模型在測試集上的表現,也能夠輸出每一步的分數
只須要將
model = XGBClassifier()
model.fit(X_train, y_train)
變爲:
model = XGBClassifier() eval_set = [(X_test, y_test)] model.fit(X_train, y_train, early_stopping_rounds=10, eval_metric="logloss", eval_set=eval_set, verbose=True)
那麼它會在每加入一顆樹後打印出 logloss
[31] validation_0-logloss:0.487867 [32] validation_0-logloss:0.487297 [33] validation_0-logloss:0.487562
並打印出 Early Stopping 的點:
Stopping. Best iteration:
[32] validation_0-logloss:0.487297
gradient boosting 還有一個優勢是能夠給出訓練好的模型的特徵重要性,
這樣就能夠知道哪些變量須要被保留,哪些能夠捨棄
須要引入下面兩個類
from xgboost import plot_importance from matplotlib import pyplot
和前面的代碼相比,就是在 fit 後面加入兩行畫出特徵的重要性
model.fit(X, y)
plot_importance(model)
pyplot.show()
如何調參呢,下面是三個超參數的通常實踐最佳值,能夠先將它們設定爲這個範圍,而後畫出 learning curves,再調解參數找到最佳模型:
接下來咱們用 GridSearchCV 來進行調參會更方便一些:
能夠調的超參數組合有:
樹的個數和大小 (n_estimators and max_depth)
.
學習率和樹的個數 (learning_rate and n_estimators)
.
行列的 subsampling rates (subsample, colsample_bytree and colsample_bylevel)
.
下面以學習率爲例:
先引入這兩個類
from sklearn.model_selection import GridSearchCV from sklearn.model_selection import StratifiedKFold
設定要調節的 learning_rate = [0.0001, 0.001, 0.01, 0.1, 0.2, 0.3]
和原代碼相比就是在 model 後面加上 grid search 這幾行:
model = XGBClassifier() learning_rate = [0.0001, 0.001, 0.01, 0.1, 0.2, 0.3] param_grid = dict(learning_rate=learning_rate) kfold = StratifiedKFold(n_splits=10, shuffle=True, random_state=7) grid_search = GridSearchCV(model, param_grid, scoring="neg_log_loss", n_jobs=-1, cv=kfold) grid_result = grid_search.fit(X, Y)
最後會給出最佳的學習率爲 0.1
Best: -0.483013 using {'learning_rate': 0.1}
print("Best: %f using %s" % (grid_result.best_score_, grid_result.best_params_))
咱們還能夠用下面的代碼打印出每個學習率對應的分數:
means = grid_result.cv_results_['mean_test_score'] stds = grid_result.cv_results_['std_test_score'] params = grid_result.cv_results_['params'] for mean, stdev, param in zip(means, stds, params): print("%f (%f) with: %r" % (mean, stdev, param)) -0.689650 (0.000242) with: {'learning_rate': 0.0001} -0.661274 (0.001954) with: {'learning_rate': 0.001} -0.530747 (0.022961) with: {'learning_rate': 0.01} -0.483013 (0.060755) with: {'learning_rate': 0.1} -0.515440 (0.068974) with: {'learning_rate': 0.2} -0.557315 (0.081738) with: {'learning_rate': 0.3}
最後附上完整的代碼
# coding=utf-8
from numpy import loadtxt from xgboost import XGBClassifier from sklearn.model_selection import train_test_split from sklearn.metrics import accuracy_score from xgboost import plot_importance from matplotlib import pyplot from sklearn.model_selection import GridSearchCV from sklearn.model_selection import StratifiedKFold dataset = loadtxt('pima-indians-diabetes.csv', delimiter=",") X = dataset[:, 0:8] Y = dataset[:, 8] seed = 7 test_size = 0.33 X_train, X_test, y_train, y_test = train_test_split(X, Y, test_size=test_size, random_state=seed) model = XGBClassifier() learning_rate = [0.0001, 0.001, 0.01, 0.1, 0.2, 0.3] param_grid = dict(learning_rate=learning_rate) kfold = StratifiedKFold(n_splits=10, shuffle=True, random_state=7) grid_search = GridSearchCV(model, param_grid, scoring="neg_log_loss", n_jobs=-1, cv=kfold) grid_result = grid_search.fit(X, Y) eval_set = [(X_test, y_test)] model.fit(X_train, y_train, early_stopping_rounds=10, eval_metric="logloss", eval_set=eval_set, verbose=True) # plot_importance(model) # pyplot.show() y_pred = model.predict(X_test) predictions = [round(value) for value in y_pred] accuracy = accuracy_score(y_test, predictions) print("Accuracy: %.2f%%" % (accuracy * 100.0)) #最佳的學習率 print("Best: %f using %s" % (grid_result.best_score_, grid_result.best_params_)) # 打印出每個學習率對應的分數 means = grid_result.cv_results_['mean_test_score'] stds = grid_result.cv_results_['std_test_score'] params = grid_result.cv_results_['params'] for mean, stdev, param in zip(means, stds, params): print("%f (%f) with: %r" % (mean, stdev, param))
轉載連接:https://www.jianshu.com/p/7e0e2d66b3d4