import numpy as np
import math
import pandas as pd

from sklearn.cross_validation import train_test_split
from sklearn.ensemble import GradientBoostingClassifier
from sklearn.grid_search import GridSearchCV
from sklearn.ensemble import GradientBoostingClassifier

def score_model(model,datax,datay):
    class_index={}
    for i,x in enumerate(model.classes_):
        class_index[x]=i
    score_matrix=model.predict_proba(datax)
    return sum(math.log(max(min(scores[class_index[datay[i]]],1-1e-15),1e-15)) for i,scores in enumerate(score_matrix))/len(datay)

if __name__ == '__main__':
    data=pd.read_csv(r'D:\Kaggle\ottogroup\train.csv')
    toscore=pd.read_csv(r'D:\Kaggle\ottogroup\test.csv')
    xcolumns=[x for x in data.columns if x.startswith('feat_')]
    trainx,trainy=data[xcolumns],data.target
    paramlist=[{'learning_rate':[.15],'max_depth':[7,10],'n_estimators':[100],'subsample':[0.5,0.85,1]}]
    grid_gbm=GridSearchCV(GradientBoostingClassifier(),paramlist,scoring=score_model,verbose=1,n_jobs=2)
    grid_gbm.fit(trainx,trainy)
    print(grid_gbm.best_params_)
    print(grid_gbm.best_score_)