Skip to content
Quantoros

Which LightGBM settings can help when training exhausts RAM?

Classic MLLevel: IntermediateBoosting memory budget

Which LightGBM settings can help when training exhausts RAM?

Hints
  • State the task, inputs and prediction target.
  • Compare with a simple baseline: assumptions, objective and independent validation.

Answer

The FAQ recommends bounding histogram_pool_size and reducing num_leaves or max_bin. Histogram cache memory is only part of total usage: the dataset and other allocations remain, so histogram_pool_size is not a hard process-memory limit. Fewer leaves or bins may also change model quality. Measure peak RSS and validation performance, and consider the cost of multiple concurrent fits during parameter search instead of budgeting only one model.

Source and licence: LightGBM developers — Official FAQ · MIT · answer: Quantoros editorial

Related questions