Skip to content
Quantoros

Why can forcing row-wise LightGBM histograms cause memory pressure although it improves training speed?

Classic MLLevel: IntermediateLightGBM · Parameters

Why can forcing row-wise LightGBM histograms cause memory pressure although it improves training speed?

Hints
  • Algorithm layout can duplicate intermediate storage.
  • Optimize for the actual resource bottleneck.

Answer

The documented row-wise mode increases Dataset memory, potentially about doubling that part of storage. Benchmark memory as well as time, and consider column-wise mode when RAM is the binding constraint.

Source and licence: LightGBM · Parameters · MIT · answer: Quantoros editorial

Related questions