Make it so that all internal results loading use the SamplesSummary object instead of the Samples object wherever possible.
This has two benefits:
-Loading samples.csv / search_internal.dill in order to load results when resuming a model-fit was very slow.
Recent updates to output.yaml mean autofit can now run without outputting a samples.csv file or search_internal.dill file at all (to save hard disk space). The samples summary file cannot be customized to not be output, meaning the information for resuming runs is always available.
This is motivated for search chaining prior passing, which by using the SamplesSummary resumes previous fits much faster.
Make it so that all internal results loading use the
SamplesSummary
object instead of theSamples
object wherever possible.This has two benefits:
-Loading
samples.csv
/search_internal.dill
in order to load results when resuming a model-fit was very slow.output.yaml
mean autofit can now run without outputting asamples.csv
file orsearch_internal.dill
file at all (to save hard disk space). The samples summary file cannot be customized to not be output, meaning the information for resuming runs is always available.This is motivated for search chaining prior passing, which by using the
SamplesSummary
resumes previous fits much faster.