Result

Result: what a run returns — predictions, settlements, and diagnostics.

The vectorbt lesson: a backtest should hand back a rich object you can interrogate (.stats(), .score, .rank_against(problem)), not a dict of floats. The leaderboard and rebase-hillclimb’s val_score: line both read from here.

class emflow.run.result.Result(problem: 'str', model: 'str', objective: 'str', lower_is_better: 'bool', split: 'str', mode: 'str', predictions: 'pd.DataFrame', settlements: 'pd.DataFrame', analysis: 't.Dict[str, dict]' = <factory>, aggregate: 'str' = 'mean')

Bases: object

problem: str

model: str

objective: str

lower_is_better: bool

split: str

mode: str

predictions: DataFrame

settlements: DataFrame

analysis: Dict[str, dict]

aggregate: str = 'mean'

property score: float

Overall objective score: per-origin scores pooled by the number of scored timestamps for mean-type metrics, summed for sum-type ones (e.g. trading revenue).

property n_scored: int

property n_origins: int

stats() → dict

to_leaderboard_row() → dict

rank_against(problem: Problem) → t.Optional[int]

Position this score would have taken on the problem’s published leaderboard (None if the problem has no reference scores).