If I'm not overlooking anything, there is no convenient way to access predictions that are stored in a run. The run object only exposes functions which use the data internally (e.g. get_metric_fn). Internally predictions are downloaded and processed (here).
I propose we add a predictions property which can serve the predictions in dataframe format through lazy loading. The remainder of the code should then also be refactored to use this property.
If I'm not overlooking anything, there is no convenient way to access predictions that are stored in a run. The
runobject only exposes functions which use the data internally (e.g.get_metric_fn). Internally predictions are downloaded and processed (here).I propose we add a
predictionsproperty which can serve the predictions in dataframe format through lazy loading. The remainder of the code should then also be refactored to use this property.