I am working on a problem that has time-series sales data from multiple stores and I want to use a TFP STS model to model the seasonality in the data, but I would like to pool the seasonality parameters across stores using a multilevel model.
Does anyone know if this is possible with TFP or have ideas about how to implement the pooling?
Thanks
Hi @Brendan_Doner,
You might have a solution by this time. Here are few more pointers. As of now, TFP lacks an integrated multilevel STS model. However, Tensorflow Probability STS API can be used to achieve parameter pooling across stores in a multilayer model using the concept sharing parameter across different stores in the model. Please use this multilevel_modeling tutorial as a reference and adapt to your use case.
Thank You