loo
loo copied to clipboard
stan-dev
loo R package for approximate leave-one-out cross-validation (LOO-CV) and Pareto smoothed importance sampling (PSIS)
In some cases models are specified ("compressed") with frequency weights to speed up the fitting, however, this doesn't work well with loo as the Pareto k's indicates that all observations...
If the list of loo objects has some named elements the output column names behave differently. For example, ``` > loo_compare(list(loo1,`Model f1+f2`=loo2)) elpd_diff se_diff Model f1+f2 0.0 0.0 model1 -530.0...
This is related to #82 but should be it’s own issue. Currently we rely on users to use the model weights to create the appropriate mixture of predictive distributions. We...
This was briefly discussed in #130 , but I am making it a separate issue so that the discussion does not get lost. I will copy what was discussed there:...
Section 4.3 of https://avehtari.github.io/modelselection/diabetes.html demonstrates how to use loo package with other loss/utility function than log score. LOO-R^2 is another example. Both of these are just for point predictions, which...
The PR introduces a `subset()` method for `"psis"` objects. This is useful if you want to drop some of the contents of a `"psis"` object without breaking it. For example,...
Right now to get the pointwise values for elpd_loo, p_loo, etc., you have to do this: ```r object$pointwise[, "elpd_loo"] ``` but it would be nice if we offered a convenience...
We should have a plotting method (in loo or bayesplot) that creates the pointwise elpd difference plot from Figure 10a of [Visualization in Bayesian Workflow](https://arxiv.org/abs/1709.01449)
stacking_weights() optimize the weights, which is a (K-1) simplex. Currently, it renders the constrained optimization and optimizes the first (K-1) weights in a constraint space. When the number of models...
I am running two brms models with longitudinal time data. I want to compare the models but using the loo compare function will take away individual datapoints that are not...
