automl
Synthesizing longitudinal data with tabPFN-extensions
Hello, I am writing to ask if the extension for generating synthetic data also includes longitudinal data. Has TabPFN been trained on this type of data as well? I don't see anything that specifically handles longitudinal data, so I suppose I would have to pre-process it so that all time records for the same subject are on the same row, but I still wonder if it is possible to detect temporal patterns. Any suggestions on this?
Hi @fraspecial, we show in https://github.com/PriorLabs/tabpfn-time-series/tree/main that time-series data (maybe similar to yours) can be very well handled by our model. By transforming timestamps as seen in that package our model can handle time data quite well. We have outstanding work to transfer timestamps to the classification & regression case as well. Am I answering your question right here?
Thank you and sorry for the delay in responding. This partially answers my question, which was more focused on the generation of synthetic data rather than inference. I was interested in understanding whether the extension of synthetic data was also capable of generating longitudinal data. If I understand correctly, tabpfn-time-series only works to predict the evolution of a single target variable. But I imagine there is the possibility of integrating the two, similar to how TapPFNRegressor is used for synthesis...