machine learning in coding（python）：使用xgboost构建预测模型

标签：scikit-learn   机器学习   xgboost   machine learning in   

接上篇：http://blog.csdn.net/mmc2015/article/details/47304591

def xgboost_pred(train,labels,test):
	params = {}
	params["objective"] = "reg:linear"
	params["eta"] = 0.005
	params["min_child_weight"] = 6
	params["subsample"] = 0.7
	params["colsample_bytree"] = 0.7
	params["scale_pos_weight"] = 1
	params["silent"] = 1
	params["max_depth"] = 9
    
    
	plst = list(params.items())

	#Using 5000 rows for early stopping. 
	offset = 4000

	num_rounds = 10000
	xgtest = xgb.DMatrix(test)

	#create a train and validation dmatrices 
	xgtrain = xgb.DMatrix(train[offset:,:], label=labels[offset:])
	xgval = xgb.DMatrix(train[:offset,:], label=labels[:offset])

	#train using early stopping and predict
	watchlist = [(xgtrain, 'train'),(xgval, 'val')]
	model = xgb.train(plst, xgtrain, num_rounds, watchlist, early_stopping_rounds=120)
	preds1 = model.predict(xgtest,ntree_limit=model.best_iteration)


	#reverse train and labels and use different 5k for early stopping. 
	# this adds very little to the score but it is an option if you are concerned about using all the data. 
	train = train[::-1,:]
	labels = np.log(labels[::-1])

	xgtrain = xgb.DMatrix(train[offset:,:], label=labels[offset:])
	xgval = xgb.DMatrix(train[:offset,:], label=labels[:offset])

	watchlist = [(xgtrain, 'train'),(xgval, 'val')]
	model = xgb.train(plst, xgtrain, num_rounds, watchlist, early_stopping_rounds=120)
	preds2 = model.predict(xgtest,ntree_limit=model.best_iteration)


	#combine predictions
	#since the metric only cares about relative rank we don't need to average
	preds = (preds1)*1.4 + (preds2)*8.6
	return preds

代码具体分析有时间写，欢迎吐槽。。。。

machine learning in coding（python）：使用xgboost构建预测模型

标签：scikit-learn   机器学习   xgboost   machine learning in   

原文地址：http://blog.csdn.net/mmc2015/article/details/47304779