Benjamin T. Vincent
Benjamin T. Vincent
At the moment we have a `treatment_time` parameter. This assumes there is a treatment time and treatment continues forever after. But this is not general enough. To make this more...
Add the following metrics: - [ ] Average Estimated Treatment Effect (ATT). The average lift in the treatment period. **NOTE:** I already did this [here](https://causalpy.readthedocs.io/en/latest/notebooks/sc2_pymc.html) but I called it the...
So far I've exclusively been working in notebooks, and the plotting is fine. But when running the Quickstart in an ipython session then the plots do not appear.
Add ANCOVA as a new capability of the package - [x] Glossary: add post-test only and pretest-posttest variants to the nonequivalent group design section - [x] Flesh out with more...
Suggestion from @ricardoV94: In situations where there are multiple valid models, then we either have to pick what model we want to use, or we can do Bayesian model averaging....
Need to provide quantitative outputs/reports for synthetic control and interrupted time series. The [Causal Impact](https://google.github.io/CausalImpact/CausalImpact.html) package provides these summary stats:  For the frequentist version:...
Add an example based on my COVID-19 example here https://www.pymc.io/projects/examples/en/latest/causal_inference/excess_deaths.html - [x] add first iteration of the example - [ ] once #22 is finished, then run this NOT on...
For Bayesian synthetic control: 1. How does the prior impact sampling problems? 2. How does the prior impact the posterior estimates? [similar to #45] Do this after #22 Tagging @juanitorduz
I've experienced clearly sub-optimal weightings when running the the `WeightedProportion` custom scikit-learn model. It is likely due to bad optimisation, perhaps getting stuck by local optima. So we need to...
Check how we call `pm.sample_posterior_predictive` in `ModelBuilder.predict`. https://github.com/pymc-labs/CausalPy/blob/e011c9de204d2b3fbb8d31480faa11d53553956d/causalpy/pymc_models.py#L32-L39 More specifically, * Do we need to use the kwarg `predictions=True`? * And should we store these predictions in `idata`, rather than...