Active Transfer Learning

Ideas for kickstarting your automation

We use embeddings of pre-trained models as the basis for our Active Transfer Learning. This way you can build lightweight classifiers and extractors on top of them.

For classifiers, we recommend using Scikit Learn, the standard library for core machine learning algorithms. You can even use concepts like grid search if you implement them in thefit method. For a decision tree, for example, this might look like this:

from sklearn.tree import DecisionTreeClassifier

class ActiveDecisionTree(BaseClassifier):

    def __init__(self):
        params = {
            "criterion": ["gini", "entropy"],
            "max_depth": [5, 10, None]
        }
        self.model = GridSearchCV(DecisionTreeClassifier(), params, cv=3)
    @inputs(
        embedding_name = "distilbert-base-uncased",
        label_type = "manual"
    )
    def fit(self, embeddings, labels):
        self.model.fit(embeddings, labels)

For extractors, you can use Sequence Learn, an open-source implementation of Scikit Learn-like taggers built with Keras.


Did this page help you?