train parameters

[cellimnet]

Hello:
(1)When training, i found it cannot work when i set auc_roc as the ref_metric.
(2)What model_name do you finally use or recommand?
Thanks for your answering.

[trislaz]

Hello,
(1) Are you performing a multilabel classification ?
(2) I would recommend using the default i.e. mhmc_layers with n_layers_classif = 3. ensembling seem to have its importance also, i found the sweet spot that optimizes computation time and performances is around rep=3, n_ensemble=3.
Do not hesitate to increase nb_tiles also!

[cellimnet]

Hello:
(1) No, I just want to see the impact of different metrics on performance.
(2) Thank you for your valuable advice!

[trislaz]

Hello, 1) ok I'll look into this issue on monday!
2) you are welcome !