I understand from a previous post to this forum that total effect contributions are calculated as derivatives. So, for example, on this page


are the new means, after setting the evidence, calculated as the original mean from the data plus the estimated total effects, calculated from the derivatives? So it is not, the weighted average of the new target distribution using the midpoints of the discretization categories?

As a related question, I know I can export the posterior probability distribution of the target node from a fitted model, but I was wondering how I could get a predicted target value (actual value, not the target category) from that. I don't think I can? I was hoping to calculate a pseudo-Rsquared by comparing actual to predicted values (is there a way to do that?).
Quote 0 0
As per Total Effect Contributions, there is no such feature in BayesiaLab. The link you gave describe how contributions are computed with counterfactuals. Is it what you are interested in?

Here is a link that describes how the Total and Direct Effects are computed: https://library.bayesia.com/display/FAQ/Total+Effects+and+Direct+Effects+Calculation

As per the predicted target value, you can use Inference | Batch Inference | Posterior Probabilities. This will give you, for each row, the posterior probability distribution of your target node and its expected value. Note that you also have the R2 in Analysis | Network Performance | Target.

Hope this helps
Quote 0 0