XGBoost only outputting two classes

Pump it Up: Data Mining the Water Table
1

Essentially I have the same problem as this person. https://discuss.analyticsvidhya.com/t/how-to-resolve-multi-class-prediction-error-in-xgboost-in-r/7030

table(target)
    0     1     2 
22824  4317 32259 

table(pred)
    0     1 
13559  1291

As you can see XGBoost doesn’t output three classes.

Here is my code.

best_params <- list(‘max.depth’ = 2,
‘eta’ = 0.010,
‘gamma’ = 1,
‘colsample_bytree’ = 0.5,
‘min_child_weight’ = 2,
‘objective’ = “multi:softmax”,
‘num_class’ = 3,
‘eval_metric’ = ‘merror’
)

model <- xgboost(train1, target, params=best_params, nrounds=100)

pred <- predict(model, test1)

Any idea where I am going wrong here?

3

I managed to find the solution:

target <- as.numeric(train$status_group)
1 Like
4

Hi @kamchatang, here

you can find our approach (see also the pdf).

we have classified every row in a binary target: to repare / no repare.