randomForest：如何控制袋内/袋外样品的比例？

Question

I'm using R randomForest for various regression tasks. The hyperparameter tuning is still mysterious to me. I've got a handle on tuning ntree and mtry , but it makes intuitive sense that I'd want to also tune the number of samples in each bag as an additional balance of model bias & variance.

Based on the documentation, I thought that this is what sampsize does. But reading the function arguments reveals that it's more complicated than that. If I run with replacement ( replace = TRUE ), it seems I have no control over the proportion of in-bag / out-of-bag samples. In fact, with replace = TRUE , I don't think the proportion that the algorithm uses is even documented.

Documentation : sampsize: Size(s) of sample to draw.

Function arguments : sampsize = if (replace) nrow(x) else ceiling(.632*nrow(x))

Is there a way to control the proportion of in-bag-samples? Is this even a worthwhile tuning parameter?

Answer 1

The parameter sampsize = if (replace) nrow(x) else ceiling(.632*nrow(x)) only means that its default value is nrow(x) when replace is true and ceiling(.632*nrow(x)) otherwise.

However, you can change as you wish by assigning a value to it:

randomForest(X, replace = T, sampsize = 10)