Accuracy Calculator: How are NMI and AMI calculated?

[HSinger04]

Hello, I have two questions regarding the calculation of the NMI / AMI metrics: I used tester.test (where tester is an instance of GlobalEmbeddingSpaceTester) with splits_to_eval=[('test', ['samples'])] as one of the arguments.

1. Is my understanding correct that the NMI / AMI metrics get calculated by using the real labels of the test split and labels predicted via k-means-clustering on the test split?
2. And as such, information from samples also doesn't get used at all?

Thank you in advance!

[KevinMusgrave]

Yes to both of your questions