Reproducibility

[PabloMeseguerEsbri]

Can you provide the hyperparemetres for the reproducibility of few-shot learning for TCGA Breast of MIL models?

[guillaumejaume]

Apologies for the late reply.

In linear probing, we used logistic regression with sklearn default L2 regularization (set to 1.0) and lbfgs solver. We set the maximum number of training iterations to 10,000 for all experiments.

In prototyping, there is no hyperpameter.

Let me know if that answers your question.

[PabloMeseguerEsbri]

For sure, thanks!