Performance settings¶
n_jobs¶
Description¶
The number of threads to use during the training and prediction.
Optimizes the speed of execution. This parameter doesn't affect results: a model is bit-identical whatever the number of threads.
Possible values:
None— Use all processor cores.- A positive integer — Use that many threads.
- A negative integer \(n\) — Use \(cpu\_count + 1 + n\) threads, so
-1uses all cores and-2all but one.
Type
int
Default value
None (the number of threads is equal to the number of processor cores)
hist_precision¶
Description¶
The precision of the gradient histograms used to search for splits.
Possible values:
full— Exact, full-precision accumulation.quantized— Faster accumulation in quantized integers.
Type
string
Default value
None (full)
refine_closed_form_tier2¶
Description¶
Calculate the leaf refinement steps (see
leaf_refine_steps) of a poisson model in closed form instead
of with an exact line search. This is faster, and the leaf values differ from the exact path by
about \(10^{-7}\). False uses the exact path.
Type
bool
Default value
True
incremental_mu¶
Description¶
Update the predicted mean \(\mu = e^{a}\) of a poisson model incrementally from one iteration to
the next instead of recalculating it. This is faster, and the predictions differ from the exact
path by about \(10^{-9}\) to \(10^{-6}\).
Type
bool
Default value
False