Skip to content

Objectives

MPSBoost currently supports these objectives:

  • squared error
  • binary logistic
  • native CPU multiclass softmax
  • quantile
  • Poisson
  • Tweedie

Parameter entry

Regression estimators choose advanced objectives through loss:

model = mb.GradientBoostingRegressor(loss="quantile", quantile_alpha=0.9)
model = mb.GradientBoostingRegressor(loss="poisson")
model = mb.GradientBoostingRegressor(loss="tweedie", tweedie_variance_power=1.5)

Model persistence

Advanced objective parameters are written into the native model payload. Model loading validates that estimator parameters and model objectives are compatible, preventing predictions from being interpreted under the wrong objective.