1
82%
main: 83%

Ran 20 Jul 2026 05:23PM UTC

Files 169

Run time 7s

Badge

Embed ▾

Committed 20 Jul 2026 05:04PM UTC coverage: 81.626% (+0.02%) from 81.607%

Job # 29762333840.1

Build Type

push

github

Committed by

Flix6x

Commit Message

Expose model-params in the forecasting pipeline config

The LightGBM configuration was a hard-coded constant, so tuning it meant
editing library source. Add a `model-params` config field that is merged over
the defaults, so callers override only the keys they name.

Overrides reach Darts' LightGBMModel, which means besides LightGBM's own
parameters they also reach `add_encoders` and `categorical_future_covariates`.
`lags`, `lags_future_covariates` and `output_chunk_shift` stay derived per
forecast horizon and are not overridable.

This matters for holiday handling in particular: `min_child_samples` defaults
to 50 while a 73-day training window holds only 2-5 holidays, so the binary
annotation regressor cannot be split on at all. Making that tunable is a
prerequisite for evaluating any fix.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01E6t3NeRRhFtLc4KvCCxFBt
Signed-off-by: F.N. Claessen <felix@seita.nl>

Coverage Stats

14900 of 18254 relevant lines covered (81.63%)

0.82 hits per line

FlexMeasures / flexmeasures / 29762333840 / 1
82%
main: 83%

README BADGES
x

Markdown

Textile

RDoc

HTML

Rst

Source Files on job 29762333840.1

FlexMeasures / flexmeasures / 29762333840 / 1 82% main: 83%

README BADGES x

Markdown

Textile

RDoc

HTML

Rst

Source Files on job 29762333840.1

FlexMeasures / flexmeasures / 29762333840 / 1
82%
main: 83%

README BADGES
x