Hands-On Artificial Intelligence for IoT

Bagging and pasting

In voting, we used different algorithms for training on the same dataset. We can also achieve ensemble learning by using different models with the same learning algorithm, but we train them on different training data subsets. The training subset is sampled randomly. The sampling can be done with replacement (bagging) or without replacement (pasting):

• Bagging: In it, additional data for training is generated from the original dataset using combinations with repetitions. This helps in decreasing the variance of different models.
• Pasting: Since pasting is without replacement, each subset of the training data can be used at most once. It's more suitable if the original dataset is large.

The scikit library has a method for ...