sudeepy
commented
2 years ago Per @shorowit, might be the same issue as https://github.com/NREL/buildstockbatch/issues/276.
Closed sudeepy closed 2 years ago
sudeepy
commented
2 years ago Per @shorowit, might be the same issue as https://github.com/NREL/buildstockbatch/issues/276.
nmerket
commented
2 years ago @sudeepy This is a "feature" of the downselect sampling. See the second paragraph here. You'll see in your log file that you're sampling twice: Once at a fairly large number of datapoints and then again to get your final sample. The point of this is to get your downselected sample to have n_datapoints number of datapoints. It does a large sample, downselects, figures out the ratio of shrinkage from the downselect, does a second sample followed by a downselect to get the number of datapoints you actually want. Setting resample: false will just do one sample at n_datapoints and then downselect based on your criteria, but managing the final number of datapoints will be up to you. We should probably just remove this feature because it confuses people.
nmerket
commented
2 years ago So just to be clear, this isn't the same as #276.
Describe the bug I'm running a vanilla checkout of
buildstockbatchandresstock, using downselect logic to simulate a single datapoint in a specific county.When using
run_workflowinresstock, doing a single datapoint simulation takes a few minutes, but running throughbuildstockbatchtakes 30 min. 20 of those minutes is spent on downselect logic, while the remaining 10 is seemingly on the simulation. This happens both on initial run and repeated runs. I have a local copy of the weather archive that my project file points to.To Reproduce Steps to reproduce the behavior:
sampler: type: residential_quota_downselect args: n_datapoints: 1 logic:
workflow_generator: type: residential_default args: timeseries_csv_export: reporting_frequency: Hourly include_enduse_subcategories: true
baseline: n_buildings_represented: 133172057 # Total number of residential dwelling units in contiguous United States, including unoccupied units, resulting from a census tract level query of ACS 5-yr 2016 (i.e. 2012-2016), using this script: https://github.com/NREL/resstock-estimation/blob/master/sources/spatial/tsv_maker.py.
upgrades:
apply_logic:
costs:
Expected behavior I'm new to
resstockandbuildstockbatch, but based on my discussion in the resstock repo with @joseph-robertson, 30 minutes to run a single simulation sounds high, especially since the same simulation (minus the downselect logic) viarun_workflowtakes under 1 minute.Logs
Please copy pertinent sections of log files here (e.g.
sampling.out,job.out-*,postprocessing.out,sinularity-output.log, etc.)(side note: where can I find those other log files?)
Platform (please complete the following information):
Additional context None