FlexMeasures / flexmeasures / 37748286887

87%

push

github

Save forecasts like any other data, dropping beliefs that repeat the one before (#2688)

* data/services: drop forecasts that repeat the belief right before them

Context:
- Forecasts were saved with save_changed_beliefs_only=False since #1546. The
  reason, which Belhsan recalled in #2685, was gaps in multi-viewpoint
  forecasts: dropping unchanged beliefs mishandled values that change back
  within one save. #2687 fixed that.
- His other reason, that a viewpoint should keep its full horizon for
  lookups by exact belief time or horizon, does not hold up: nothing in
  FlexMeasures looks forecasts up by exact belief time, a horizons_at_least
  filter (the API's horizon) still finds the earlier, equal belief, and only
  an opt-in horizons_at_most in a reporter's input could see fewer.
- Meanwhile False made a forecast re-run from the same viewpoint, as an
  explicit start or an automation window repeats it, fail on the unique
  constraint, and stored every confirming forecast again.

Change:
- Save forecasts with the default, dropping beliefs that repeat the belief
  right before them. A re-run from the same viewpoint now records nothing new
  when its values are the same, and leaves different values to
  FLEXMEASURES_ALLOW_DATA_OVERWRITE, like any other data.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0129WrXeJ5gia2pctFH93BqC
Signed-off-by: F.N. Claessen <claessen@seita.nl>

* cli/tests: pin re-running and confirming forecasts

Context:
- Saving forecasts with the default dedupe changes two things a user sees,
  and neither was tested.

Change:
- Running the same forecast twice from an explicit start succeeds and records
  it once. On main the second run fails with a UniqueViolation.
- A new run that forecasts the same values adds no beliefs, while a lookup of
  the most recent beliefs after it still returns the whole forecast. On main
  the second run stores all 48 beliefs again.

Co-Author... (continued)

21 of 23 new or added lines in 2 files covered. (91.3%)

21196 of 24510 relevant lines covered (86.48%)

0.86 hits per line