At the ILAMB 2025 workshop, from talking with several users of ILAMB, I realized that we may like to have many kinds of meta analyses some of which wrap around lower levels of the hierarchy, say at the "variable" level. Some ideas that emerged:
- Create a pairwise plots of all containing reference data sources. We would have to think if and how these could be integrated into the unified dashboard.
- Cross-evaluate data sources against each other using ILAMB and mark (somehow) model scores which are within this spread (a la Erwin, AMBER framework). This gives some context (uncertainty) of how big score differences really are.
- Extract models and reference data and perform PCA, showing dominant models (a la Anthony's poster)
Some of these could be their own metric, but others are clearly post-processing of results ILAMB already generates.
At the ILAMB 2025 workshop, from talking with several users of ILAMB, I realized that we may like to have many kinds of meta analyses some of which wrap around lower levels of the hierarchy, say at the "variable" level. Some ideas that emerged:
Some of these could be their own metric, but others are clearly post-processing of results ILAMB already generates.