Video: Prediction vs. InferenceΒΆ
Note: Subtitles are available (click CC button at the bottom of the video)
Note
Oh yes, when I say in the video lecture that you do not need much of the statistical theory to do prediction tasks, it is only partially true. If you want to develop best prediction methods, you do need theory! You also need it to understand the basic intuition behind the prediction toolkit. Every single method widely used in data science (e.g., Lasso, Ridge, Random forest, Boosting, etc.) comes with beautiful mathematical proofs in some statistics journals. The practitioners, though, are persuaded to use those methods not by the proofs in theorems but by the actual proven declines in prediction errors that are easy to verify.