XGBoost解决多分类问题

写在前面的话

    XGBoost官方给的二分类问题的例子是区别蘑菇有无毒，数据集和代码都可以在xgboost中的demo文件夹对应找到，我是用的Anaconda安装的XGBoost，实现起来比较容易。唯一的梗就是在终端中运行所给命令：  ../../xgboost mushroom.conf 时会报错，是路径设置的问题，所以我干脆把xgboost文件夹下的xgboost.exe拷到了mushroom.conf配置文件所在文件夹下，这样直接定位到该文件夹下就可以运行： xgboost mushroom.conf。二分类数据预处理，也就是data wraggling部分的代码有一定的借鉴意义，值得一看。

    多分类问题给的例子是根据34个特征识别6种皮肤病，由于终端中运行runexp.sh没有反应，也不报错，所以我干脆把数据集下载到对应的demo文件夹下了,主要的代码如下，原来有部分比较难懂的语句我自己加了一些注释，这样理解起来就会顺畅多了。

#! /usr/bin/pythonimport numpy as npimport xgboost as xgb# label need to be 0 to num_class -1# if col 33 is '?' let it be 1 else 0, col 34 substract 1data = np.loadtxt('./dermatology.data', delimiter=',',converters={33: lambda x:int(x == '?'), 34: lambda x:int(x)-1 } )sz = data.shapetrain = data[:int(sz[0] * 0.7), :] # take row 1-256 as training settest = data[int(sz[0] * 0.7):, :]  # take row 257-366 as testing settrain_X = train[:,0:33]train_Y = train[:, 34]test_X = test[:,0:33]test_Y = test[:, 34]xg_train = xgb.DMatrix( train_X, label=train_Y)xg_test = xgb.DMatrix(test_X, label=test_Y)# setup parameters for xgboostparam = {}# use softmax multi-class classificationparam['objective'] = 'multi:softmax'# scale weight of positive examplesparam['eta'] = 0.1param['max_depth'] = 6param['silent'] = 1param['nthread'] = 4param['num_class'] = 6watchlist = [ (xg_train,'train'), (xg_test, 'test') ]num_round = 5bst = xgb.train(param, xg_train, num_round, watchlist );# get predictionpred = bst.predict( xg_test );print ('predicting, classification error=%f' % (sum( int(pred[i]) != test_Y[i] for i in range(len(test_Y))) / float(len(test_Y)) ))# do the same thing again, but output probabilitiesparam['objective'] = 'multi:softprob'bst = xgb.train(param, xg_train, num_round, watchlist );# Note: this convention has been changed since xgboost-unity# get prediction, this is in 1D array, need reshape to (ndata, nclass)yprob = bst.predict( xg_test ).reshape( test_Y.shape[0], 6 )ylabel = np.argmax(yprob, axis=1)  # return the index of the biggest proprint ('predicting, classification error=%f' % (sum( int(ylabel[i]) != test_Y[i] for i in range(len(test_Y))) / float(len(test_Y)) ))

    结果如下：

[0]train-merror:0.011719test-merror:0.127273[1]train-merror:0.015625test-merror:0.127273[2]train-merror:0.011719test-merror:0.109091[3]train-merror:0.007812test-merror:0.081818[4]train-merror:0.007812test-merror:0.090909predicting, classification error=0.090909[0]train-merror:0.011719test-merror:0.127273[1]train-merror:0.015625test-merror:0.127273[2]train-merror:0.011719test-merror:0.109091[3]train-merror:0.007812test-merror:0.081818[4]train-merror:0.007812test-merror:0.090909predicting, classification error=0.090909

    

    不管是直接返回诊断类型，还是返回各类型的概率，然后取概率最大的那个对应的类型的index，结果都是一样的。

结语

    强烈建议大家使用python notebook来实现代码，当有不明白的代码时看一下执行后的结果能帮助我们很快理解。同时要感叹一下，看大神们的代码感觉好牛X，对我这个XGBoost paper看过两遍还没能完全领略算法精髓的人来说只能拿来主义了，希望后面有机会去读一读算法源码。