🔊 Warn user when a model is reused for the the latent task

[enryH]

In case of typos in configurations, the analyze latent reloads a model and finds the wrong dimensions. Should it be possible to always retrain?

Definitely log that a pre-trained model was found.

[ri-heme]

Definitely log that a pre-trained model was found.

If I'm not mistaken, this already happens. In the log file, the message: "Re-loading model" changes to "Training model" depending on whether a pre-trained model is being used. But agree that this should be explained more clearly. Perhaps also adding it as a warning? Or checking that the weights/dimensions match and the model can be ran without issue?

Should it be possible to always retrain?

The idea behind not doing this is to prevent an old model from being erased/overwritten, and also save time re-running the task if you want to just (re)generate the plots.