Experiment

Run a predictor against an emflow problem (the standard walk-forward loop).

This is the plain evaluation runner used in the example notebooks: reset the environment, train on the initial window, then for each step predict, advance the environment to reveal the targets, and score with the problem’s objective. For untrusted submissions where leakage must be impossible, use emflow.experiments.verifier.Verifier instead.

class emflow.experiments.experiment.Experiment(problem, model)

Bases: object

run()

Run the reset → train → (predict → step → score)* loop.

Returns a dict with the per-step losses and their mean.