Repository navigation
Column-wise quantiles in forecasts to reduce memory load #1203
Description
In HazardForecast and ImpactForecast, the entire sparse matrix is densified before computing the quantiles. This will easily crash for larger, sparse datasets. Instead of densifying the entire matrix, we should densify single columns, and compute the respective quantile individually. This should reduce the memory load of quantile (but can make it a bit slower).
Activity
[-]Column-wise quantiles to reduce memory load[/-][+]Column-wise quantiles in forecasts to reduce memory load[/+]perf: compute forecast quantiles block-wise to bound memory
peanutfun commentedon Sep 2, 2026
Resolved in #1313