output of evaluator.run

[vsk-phi]

Hi,
I want to use the supervised evaluator to collect the prediction results. What i found is that after collecting the state of the evaluator eg state = evaluator.run(data_loader, max_epochs=1), the state.output has y_pred,y only for the last batch.
Is there a way one can collect the output of all batches, if not necessarily in one go?
The use case is that, i want to write the output predictions, as well as the file names. I can conceivably do this separately with a simple iteration over the val_loader with nograd, however i want to know if this can be done in an evaluator run.
Thanks

[sadra-barikbin]

Hi @vsk-phi , ignite.handlers.EpochOutputStore might be the solution for you. It lets you to access model outputs being collected during an epoch, at the end of the epoch. Those are collected as a list of (y_pred, y)s. Does this work for you?

[vfdev-5]

Hi @vsk-phi , you can use https://pytorch.org/ignite/generated/ignite.handlers.stores.EpochOutputStore.html#epochoutputstore to collect all outputs.

[vfdev-5]

@sadra-barikbin is very quick :)

[vsk-phi]

Hi, Thanks both so much for super fast response. It seems that I can use this; probably I didn’t search the documentation enough ☹