Melissa for sensitivity analysis requires iterative statistics (also sometimes called online statistics our incremental statistics), i.e. algorithms that enable to update the current statistics with a new sample a s soon as this one is available. This enables to work with a reduce memory footprint and overlap stats updates and sample generations. Here is a really nice initiative on github to develop a stat library dedicated to these iterative stats (GitHub - IterativeStatistics/BasicIterativeStatistics: This package implements iterative algorithms to compute basic statistics.) with extensive validation tests (numerical stability can be tricky with these algorithms). For the moment mean variance, covariance, Sobol (several versions) are supported. More to come soon (high order moments, quantiles). We rely on these implementations for Melissa.
Related topics
| Topic | Replies | Views | Activity | |
|---|---|---|---|---|
| Iterative statistics update - [breaking change] | 0 | 50 | June 29, 2023 | |
| About the Sensitivity Analysis category | 0 | 45 | May 4, 2023 | |
| Zero division in iterative statistic computation | 1 | 58 | May 3, 2023 | |
| About the Data Assimilation category | 0 | 45 | May 4, 2023 | |
| Statistical differences noticed from ZMQ -> ADIOS2 | 3 | 35 | May 9, 2024 |