Option to pass at rgcca_predict to get class predicted probabilities

[tonyliang19]

Hello, I was testing the rgcca_predict functionality with my data on classifcation problem. I think it's very awesome to just have a string to change the underlying model to use. But I noticed the returned output is simply just the predicted class and not the probabilities itself.

So, I wonder if you have implemented a way that you just pass like an additional parameter to your rgcca_predict like a normal S3 predict(..., type = "probs"), whereas you could pass that type arg to return either probability or link function output etc. Thanks!

[GFabien]

Hi @tonyliang19!
Thank you for your message. Would you prefer a new argument type to change the prediction output into probabilities, or would a new output probs with the probabilities when relevant or empty otherwise be enough? We might be a bit far from the regular S3 predict for the 1st option. What do you think?

[tonyliang19]

Hi @tonyliang19!
Thank you for your message. Would you prefer a new argument type to change the prediction output into probabilities, or would a new output probs with the probabilities when relevant or empty otherwise be enough? We might be a bit far from the regular S3 predict for the 1st option. What do you think?

Thanks for your consideration on this matter, I would say if the first option deviates a lot from regular S3 predict, then adding a new output probs would be adequate and enough. As it complements by your other standard output of a rgcca predict.

[GFabien]

You can try the branch: https://github.com/rgcca-factory/RGCCA/tree/add_probs_in_predict; we'll merge it if it suits your needs!

[tonyliang19]

Amazing, I will try this as soon as possible and you let you know!

[tonyliang19]

You can try the branch: https://github.com/rgcca-factory/RGCCA/tree/add_probs_in_predict; we'll merge it if it suits your needs!

Thats exactly what I needed! I liked that you provided the probability for each class predicted. This works for now, although I have not tested with the case of multi-class classifications. Again, thanks for you work on this