Previously, data were not able to be disambiguated between series in a gridsearch. This was because points found by the (metrics, grid_params) specifications were pooled across series before checking for parameter matches.
E.g. for a single combo of metric and grid_param, the Kalman series may have an optimum at (3, 1, 2). The equivalent point may have an optimum of (3, 1, 6) in the Total Variation series. Depending on parameter match criteria, this could result in (3, 1, 2) and (3, 1, 6) in the Kalman series and in the Total Variation series being seen as a match (a total of four matches).
There is a hack in gen_experiments that resolves this temporarily: any use of plot_prefs needs to specify exactly one params_or dictionary for each series.
Previously, data were not able to be disambiguated between series in a gridsearch. This was because points found by the (metrics, grid_params) specifications were pooled across series before checking for parameter matches.
E.g. for a single combo of metric and grid_param, the Kalman series may have an optimum at (3, 1, 2). The equivalent point may have an optimum of (3, 1, 6) in the Total Variation series. Depending on parameter match criteria, this could result in (3, 1, 2) and (3, 1, 6) in the Kalman series and in the Total Variation series being seen as a match (a total of four matches).
There is a hack in gen_experiments that resolves this temporarily: any use of
plot_prefs
needs to specify exactly oneparams_or
dictionary for each series.