manager.trainer.should_retrain

manager.trainer.should_retrain(last_end_dev, *, max_age, force=False, now=None)

Decide whether a forecaster should be retrained.

The cadence policy lives in ConfigEntsoe.retrain_max_age. Callers pass max_age in explicitly so the policy stays in config, not in code.

Parameters

Name	Type	Description	Default
last_end_dev	Optional[pd.Timestamp]	End-of-development-window timestamp of the last trained model. `None` indicates no previous model.	required
max_age	pd.Timedelta	Maximum allowed age before retraining is required.	required
force	bool	Bypass the cadence gate and always return `True`.	`False`
now	Optional[pd.Timestamp]	Override “now” — used for deterministic tests. Defaults to `pd.Timestamp.now(tz="UTC")`.	`None`

Returns

Name	Type	Description
	bool	`True` when retraining should proceed (no previous model,
	bool	`force`, or age above `max_age`); `False` otherwise.

Examples

import pandas as pd

from spotforecast2_safe.manager.trainer import should_retrain

now = pd.Timestamp("2026-05-24 12:00", tz="UTC")
max_age = pd.Timedelta(days=7)

# No previous model -> retrain.
print(should_retrain(None, max_age=max_age, now=now))

# Recent model -> skip.
recent = now - pd.Timedelta(days=2)
print(should_retrain(recent, max_age=max_age, now=now))

# Old model -> retrain.
old = now - pd.Timedelta(days=10)
print(should_retrain(old, max_age=max_age, now=now))

True
False
True