BlockScience / aztec-gddt

Aztec Granular Design Digital Twin
Apache License 2.0
10 stars 0 forks source link

Implement Sweep over Start Time for the Censoring Series #236

Closed danlessa closed 4 months ago

danlessa commented 4 months ago

User Story> we don't have the the sweep, but we have most components. (i.e. the params, the censoring builders, the subset of data we use). what we don't have is primarily where to start the censoring series from (the entire series is hundreds of thousands of blocks), but I think a reasonable - but not implemented - way would be to pick a random start time for each run, and do lots of monte carlo?

notes: the source time-series has about 230k elements in L1 time the constructed time-series has about 10k elements in L1 time we may want to manually select a subset of the time series (eg. 10 at most, 20%)

  1. We provde a list of cherry-picked start times
  2. We set the count of time-series to build
  3. We set the Maximum % of cherry-picked start times to be included
  4. If list 1 is larger than the maximum set of cherry-picked time-series, then select top N until it fulfills that %
  5. For all the non cherry picked time series, sample the L1 start time randomly from the source ts (good enough: pseudorandom, excellent: sobol sampling)
jackhack00 commented 4 months ago

@danlessa List of block_numbers, where 'beaverbuild.org' started a spree of at least 13 slots in a row. When setting for simulation, we should start at least 200-300 block_numbers before the one listed, to make sure that we don't start every simulation directly with the shock. Index Builder block_number length_of_slots_in_a_row 132606 beaverbuild.org 19715891.0 13 61674 beaverbuild.org 19433198.0 13 64801 beaverbuild.org 19445274.0 13 65037 beaverbuild.org 19446165.0 13 66268 beaverbuild.org 19450890.0 13 107388 beaverbuild.org 19615545.0 13 106625 beaverbuild.org 19612631.0 13 106067 beaverbuild.org 19610517.0 13 105038 beaverbuild.org 19606581.0 13 104043 beaverbuild.org 19602512.0 13 94642 beaverbuild.org 19564625.0 13 103198 beaverbuild.org 19599046.0 13 117796 beaverbuild.org 19655403.0 13 103677 beaverbuild.org 19601007.0 13 103812 beaverbuild.org 19601571.0 13 114538 beaverbuild.org 19643290.0 14 60098 beaverbuild.org 19427323.0 14 94570 beaverbuild.org 19564317.0 14 77914 beaverbuild.org 19497529.0 14 103237 beaverbuild.org 19599224.0 14 107109 beaverbuild.org 19614434.0 15 68122 beaverbuild.org 19458179.0 15 60948 beaverbuild.org 19430416.0 15 104016 beaverbuild.org 19602401.0 16 89454 beaverbuild.org 19544029.0 17 72507 beaverbuild.org 19475612.0 17 113763 beaverbuild.org 19640428.0 17

jackhack00 commented 4 months ago

maybe slightly easier: Took the 12 longest sequences, and subtraced 300 from the block_number: 19427023.0 19497229.0 19564017.0 19598924.0 19642990.0 19430116.0 19457879.0 19614134.0 19602101.0 19475312.0 19543729.0 19640128.0

danlessa commented 4 months ago

Done by 706e14b