Open ricardoV94 opened 1 year ago
The Google Cloud Docs contain a guide on how to do this with a DL model:
https://cloud.google.com/ai-platform/docs/clv-prediction-with-offline-training-train
This paper proposes Gaussian Processes for CLV modeling, which should be within the capabilities of pymc
:
If there is interest in adding a spend model with covariates, the GP model is worth exploring further.
GP model sounds interesting, we could perhaps start with a developer notebook just to see it works well before implementing a packaged model?
That seems like the sensible approach.
Can someone more familiar with GP modeling in pymc
assign themselves to this? Right now my focus is on replicating all functionality of the lifetimes
library.
If we don't summarize individual transaction values, there should be much more flexibility in how to model user latent "spend", with e.g, timeseries component, glm predictors, ....
Would be nice to add a study case of such, perhaps motivating new summary/plotting/prediction functionality of the library.