[BUG] GMM model and specific category combinations

[felixbur]

Describe the bug and step to reproduce
super strange: for some combinations of categories with emodb and sklearn GMM classifier, the GMM classifier's predict() function returns labels that are outside the range (e.g. for [0:2], [0:3] is the result)

Here is a minimal nkululeko config to reproduce

[EXP]
root = ./experiments/emodb/
name = results
[DATA]
databases = ['emodb']
emodb = ./data/emodb/emodb
emodb.split_strategy = speaker_split
labels = ["disgust", "boredom", "anger"]
target = emotion
[FEATS]
type = ['os']
[MODEL]
type = gmm

Expected behavior

Desktop information

• OS: Ubuntu 20
• Nkululeko Version:
  93.10