#aa.to_csv('xgb1.csv',index=False)
xgb_model=xgb.XGBClassifier()
print('Train XGBoost model')
params = {'nthread':[4],
"objective": ["multi:softprob"],
"learning_rate": [0.3],
"max_depth": [6,15,30],
"min_child_weight":[1],
"subsample": [0.9],
"colsample_bytree": [0.97],
'seed':[0,88],
'missing':[-999],
'n_estimators':[20]
}
#watchlist=[(dtrain,'train'),(dvalid,'eval')]
bst=GridSearchCV(xgb_model,params,n_jobs=10,cv=StratifiedKFold(y,n_folds=5,shuffle=True),scoring='log_loss',verbose=1,refit=True)
bst.fit(train[features], y)
#trust your CV!
best_parameters, score, _ = max(bst.grid_scores_, key=lambda x: x[1])
print('Raw mlogloss score:', score)
for param_name in sorted(best_parameters.keys()):
print("%s: %r" % (param_name, best_parameters[param_name]))
y_pred = bst.predict_prob(test[features])
