The new interface we've converged on (#231) uses the Features table as both input to setup a model, and output to recover its results. Right now any parameter can have a value and two bounds. But for output, we will also need an uncertainty for that value, as well as the full covariance matrix. We discussed putting the latter in the meta attribute of the Features table. What about the former? In principle it is fully specified by the covariance matrix. But it would be nice to include in the table itself. We could also consider a 4-tuple (value, value_uncertainty, min, max), or just provide helpers to get the uncertainty out of the cov. matrix. Thoughts welcome.
The new interface we've converged on (#231) uses the Features table as both input to setup a model, and output to recover its results. Right now any parameter can have a value and two bounds. But for output, we will also need an uncertainty for that value, as well as the full covariance matrix. We discussed putting the latter in the
meta
attribute of the Features table. What about the former? In principle it is fully specified by the covariance matrix. But it would be nice to include in the table itself. We could also consider a 4-tuple (value, value_uncertainty, min, max), or just provide helpers to get the uncertainty out of the cov. matrix. Thoughts welcome.