factor_agent.evaluator.Evaluator.evaluate#
- Evaluator.evaluate(graphs, admit=None, tried=None)[source]#
Return one entry per graph, in order; see the module docstring.
- Parameters:
- graphslist of Graph
The batch, compiled together.
- admitcallable, optional
admit(position, metrics, ran)is called for every graph that ran, in order, before the next one is scored;ranis how many of the batch’s graphs ran. When it returns a name, the candidate joins the reference pool under it. Without it no candidate joins.- triedsequence of str, optional
A name per graph: every graph that ran joins the tried trials under its name after the batch (
tried).- The columns of the graphs that ran are kept until the next
- ``evaluate``, for ``join``: as float32 normal scores on each scored
- date’s rows, about 45 MB a graph on 3000 members over 15 years.