注意objective 和eval_metric

使用xgboost 进行learning to rank

train_dmatrix = xgb.DMatrix(train.drop(['group','排名','score','Promotion'], axis=1),label=train['score'])
test_dmatrix = xgb.DMatrix(test.drop(['group','排名','score','Promotion'], axis=1),label=test['score'])
#设置训练集的group，将训练集按group排序，设置group为该group中行的个数，而不是编号，需满足 sum(train_group) = len(train)
train_dmatrix.set_group(train_group)
test_dmatrix.set_group(test_group)

params = {'booster':'gbtree',
               'objective': 'rank:ndcg', 
               'eta': 0.1,
               'max_depth': 6,
               'silent':1,
              'verbose':10,
               'eval_metric':'ndcg@20-'}
xgb_model = xgb.train(params, train_dmatrix, num_boost_round=2000,
                           evals=[(train_dmatrix, 'train'),(test_dmatrix, 'test')],early_stopping_rounds=50)