Problem: Function runs slowly or runs out of memory.
Workarounds:
- Increase the value of NumBoostedTrees (the number of parallel boosted trees to consider during prediction).
Each vworker processes each of its assigned tree-boosting operations sequentially. When you increase the value of NumBoostedTrees, each boosting operation operates on smaller subset of input data.
- Decrease the input data footprint by minimizing the number of columns.
- Decrease the size of a tree in memory by adjusting the values of these arguments:
- MinNodeSize
- MaxDepth
- ColumnSubSampling