Dear all,
I tried to perform AgglomerativeClustering in reval on toy dataset and gaussian blobs, but the clustering algorithm failed to find any cluster. I tried to re-run AgglomerativeClustering starting from the fit method and moving backwards throughout the FindBestClustCV class. Since FindBestClustCV is a child class of RelativeValidation, I focused on the RelativeValidation class and found that the problem was in performing the Kuhn munkres algorithm in the test and rescale_score methods. The Kuhn munkres algorithm method in the utils module requires array-like structures of int32 or int as input variables, whereas the clustering labels I obtained from the train method of the RelativeValidation class were int64. Consequently, I modified line 84 in the relative_validation module (test method) as follows to convert any type of numpy array into a int32 numpy array:

bestperm = kuhn_munkres_algorithm(np.int32(classlab_ts), np.int32(clustlab_ts))

Accordingly, I also adjusted line 130 ( rescale_score_ method) in the same way:

me_ts = zero_one_loss(pred_lab, kuhn_munkres_algorithm(np.int32(pred_lab), np.int32(labts)))

I suggest modifying these code lines to overcome any issues due to the array structure type of clustering and classifier labels.