There is a bug in the probability output code. This has come about because initially, one hot encoding was used for the target features (instrument), but later we switched to using label encoding. The vote probability code, was not updated, so is producing nonsense.
There is a bug in the probability output code. This has come about because initially, one hot encoding was used for the target features (instrument), but later we switched to using label encoding. The vote probability code, was not updated, so is producing nonsense.
The code responsible is here: https://github.com/MetOffice/XBTs_classification/blob/master/classification/experiment.py#L659
in
generate_vote_probabilities
.