Question

What types of hypotheses can lazy learning algorithms represent versus eager learners?

Answer 1

In lazy learning algorithms, the model store the training data, and wait until the test data arrives. As soon as test data arrives, it makes prediction using the hypothesis of that algorithm and return us the output or prediction.

Whereas in Eager Learning, the model actually learn parameters after training itself on training data, which are capable of learning all type of features/trends from the training dataset, and when testing data comes, we can directly make predictions using that parameters only.

Example of lazy learners: KNN etc. In knn hypothesis space depends on the K-nearest neighbors, meaning that prediction value would depend on values of k-nearest neighbors. Like in case of classification we would make prediction of the class which occurs maximum in K-nearest neighbors, whereas in regression(continuous value, like house price prediction) we would make predictions by taking average of those K-nearest-neighbors.

Example of eager learners Decision Trees, Artificial Neural Networks etc. In Artificial Neural Network, we would train our model on training dataset, which would set optimal values of weights and bias of the model and when testing data arrives, we would forward propogate the neural network and make predictions using the output of the model. In case of regression problem(continuous values, house price prediction) the model would directly give the predicted value, whereas in classification we may design our model to predict probabilities of every class or class having maximum probability itself. The task in eager learning is the learning of the hypothesis space that commit to all entire instance space.