Orchestrate real ELPA-style ensemble forecasting workflows by triggering external sub-model training jobs (for example PyTorch/Prophet/TiDE/transformers), then computing ELPA online/offline weights from validation errors. Use when you need production-oriented ensemble training instead of lightweight simulation adapters.
This skill does not train toy adapters. It triggers real sub-model training commands from your own training codebases and then builds ELPA routing/weights from real validation errors.
Default model pool is intentionally larger than 4 and can be expanded freely.
assets/elpa_train_template.json).Create a config based on assets/elpa_train_template.json.
train_cmd.online or offline.python3 scripts/elpa_orchestrator.py \
--config assets/elpa_train_template.json \
--run-dir .runtime/elpa_run \
--manifest-out .runtime/elpa_run/train_manifest.json
This prints and records the commands that would run, without training.
python3 scripts/elpa_orchestrator.py \
--config /path/to/your_train_config.json \
--run-dir .runtime/elpa_run \
--manifest-out .runtime/elpa_run/train_manifest.json \
--execute
Use this only in an environment that has the required ML dependencies and hardware.
After each sub-model produces validation errors, run:
python3 scripts/elpa_integrator.py \
--config /path/to/your_integrate_config.json \
--output .runtime/elpa_run/elpa_policy.json
The output includes:
scores for each model from validation errorsonline_weights and offline_weightsbest_online_model and best_offline_modelbeta, dirty_interval, amplitude_window, mutant_epsilon)To support more models, append model blocks in your config with:
namegroup as online or offlinetrain_cmdNo script changes are needed for adding models.
scripts/elpa_orchestrator.py: real sub-model training command planner/executorscripts/elpa_integrator.py: ELPA score/weight builder from validation errorsassets/elpa_train_template.json: >4-model real training templateassets/elpa_integrate_template.json: ELPA integration templatereferences/config-schema.md: config field reference and placeholders