Add TransformedDataset class in order to enable inference for stochastic states
cweniger opened this issue · 1 comments
cweniger commented
Goal: Strategy for generating posteriors for parameters with implicit priors.
- Parameters are part of the model prediction
- Original priors and simulations are defined as before
- Dataset is the thing that provides information about priors, bounds and parameter names
- We should tweak the dataset to change the expected results
- We could have some transformed dataset?
- Overwrite: pnames, v, bound, prior
- The prior would have to be defined empirically, based on the distribution of the parameter in the training set.
Suggestion: TransformedDataset(dataset, sim_to_v, pnames, eff_prior, eff_bound)
cweniger commented
In v0.4 everything is based on implicit priors by construction. Closed