h2oai / db-benchmark

reproducible benchmark of database-like ops
https://h2oai.github.io/db-benchmark
Mozilla Public License 2.0
322 stars 85 forks source link

dask rounding different between different runs #136

Closed jangorecki closed 4 years ago

jangorecki commented 4 years ago

for same question dask happens to round number different, see 506570021.499 vs 506570021.5 below

      task solution           data                question                         unq_chk unqn_chk
1: groupby     dask G1_1e7_1e1_0_0        sum v1:v3 by id6 30001016;29998058;506570021.499        2
2: groupby     dask G1_1e7_1e1_0_0        sum v1:v3 by id6   30001016;29998058;506570021.5        2
3: groupby     dask G1_1e7_1e2_0_0 sum v3 count by id1:id6          506570021.499;10000000        2
4: groupby     dask G1_1e7_1e2_0_0 sum v3 count by id1:id6            506570021.5;10000000        2