On training the baseline script the log says that it is not able to train the model for several data instance. Is it the normal behaviour.
DEBUG:HierarchicalClassifier:_train_local_classifier() - Training local classifier for node: D21C, X_.shape: 286150, len(y): 0, n_targets: 0
DEBUG:HierarchicalClassifier:_train_local_classifier() - could not train node D21C
DEBUG:HierarchicalClassifier:_train_local_classifier() - Training local classifier for node: D21F, X_.shape: 286150, len(y): 0, n_targets: 0
DEBUG:HierarchicalClassifier:_train_local_classifier() - could not train node D21F
DEBUG:HierarchicalClassifier:_train_local_classifier() - Training local classifier for node: D21B, X_.shape: 286150, len(y): 0, n_targets: 0
DEBUG:HierarchicalClassifier:_train_local_classifier() - could not train node D21B
DEBUG:HierarchicalClassifier:_train_local_classifier() - Training local classifier for node: D21G, X_.shape: 286150, len(y): 0, n_targets: 0
DEBUG:HierarchicalClassifier:_train_local_classifier() - could not train node D21G
DEBUG:HierarchicalClassifier:_train_local_classifier() - Training local classifier for node: D21J, X_.shape: 286150, len(y): 0, n_targets: 0
DEBUG:HierarchicalClassifier:_train_local_classifier() - could not train node D21J
Yes, this is normal, for the baseline. it means there is not enough samples to train. Here, the challenge is to overcome this problem with the given hierarchy/name of the labels, or other approach, so that the algorithm can perform zero-shot classification (Task B).
Best,
Fernando