Is there a way to only access the new labelled data in dataset.label ?

[Karthik-Pandaram]

Hi I wanted to know if there is a way to access the newly labelled data alone from the dataset object in the AL loop.

for i in range(5):
  
  print("Dataset size", len(dataset))
  clf.fit(*dataset.dataset)
  predictions = clf.predict(test_x)
  print("Test Accuracy  :: ", accuracy_score(test_y, predictions))
  probs = predict(dataset.pool, clf)
  to_label = heuristic(probs)
  query_size = 10

  if len(to_label) > 0:
      dataset.label(to_label[: query_size])
      break
  else:
    break

here the to_label contains the indexes to be labelled but when I try train_x.iloc[x] the results are different from the those in the dataset.dataset[0].
Can some one explain me how the dataset.label adds new samples to the labelled pool?

[Dref360]

Hello!

In this case, the indices in to_label are relative to the unlabelled pool, not to the overall dataset.

To get the mapping from the pool to the "original" indices, you can use dataset._pool_to_oracle_index(to_label[: query_size]) before dataset.label.

Addtional infos:

• ActiveLearningDataset keeps track of what is labelled and what is not.
• ActiveNumpyDataset.dataset gives you a new array with labelled items only.
• ActiveNumpyDataset.pool gives you an array with unlabelled items only.
• ActiveNumpyDataset.label(indices) add items from the unlabelled pool to the labelled pool before adding these indices to the set of labelled items.

I hope this helps!

[Karthik-Pandaram]

Hi Thanks for the reply it is helpful !.

It would be great if we could have a method to get the newly labelled samples alone as a separate dataset, before adding to the pool.

[Dref360]

Hmm right, I think the easiest way is still dataset.pool[0][to_label[:query_size]] with ActiveNumpyArray, but I understand that it is a bit complicated.