Hi,
I just finished the video and have a doubt. In the video Andrew mentions Ensembling where he asks us to train several networks independently and then average them.
Should the same network be initiated independently and the results averaged or should we be using networks of different architectures for the purpose? What does he mean when he says ‘train several networks’ in this situation? Thanks in advance
Best regards
Prejith
When it comes to ensemble learning, both approaches are valid.
In bagging approach, we generate samples with replacement of the original data via a process called bootstraping. Then, we use multiple copies of the base learner to learn from different samples of data.
When it comes to stacking, we can have multiple base learners whose outputs are then combined to generate the final prediction (See this kaggle example)
Thank you very much for the detailed explanation and the link to kaggle.