Core50 benchmark with ICARL strategy

[gab709]

ICarl strategy returns an error with Core50 benchmark: scenario = CORe50(scenario="nc", run=9)

The error is: AttributeError: 'GenericCLScenario' object has no attribute 'n_classes_per_exp'

Other nc benchmarks belong to "NCScenario" class and this issue does not appear.

[AntonioCarta]

As a quick fix, you can easily compute n_classes_per_exp yourself by using exp.dataset.targets.

[ggraffieti]

Update: benchmarks created with the method create_multi_dataset_generic_benchmark have the same problem (no attribute n_classes_per_exp and classes_order). I did not investigate much, but seems that also benchmarks created with create_generic_benchmark_from_filelists (as CORe50) have the same problem.

Maybe is related to the generic benchmarks?

[ashok-arjun]

@ggraffieti I think it's related to how the structure of the generic benchmarks is.

I'd like to take up this issue. You can assign it to me.

How might I approach it? @AntonioCarta @ggraffieti

[lantao1234]

When i create a paths_benchmark, the function have the same problem.

AttributeError: 'GenericCLScenario' object has no attribute 'n_classes_per_exp'

[ema-marconato]

Is still an open issue? I'm facing the same problem with ICarL implementation.

[Melika-Ayoughi]

I have the same problem with ego objects benchmark and icarl. Is it still not fixed?