Supervised Machine Learning with Python

Pros of non-parametric models

Non-parametric models are able to learn some really complex relationships between your predictors and the output variable, which can make them really powerful for non-trivial modeling problems. Just like the regression sinusoidal wave we modeled in the decision trees, a lot of non-parametric models are fairly tolerant to data scale as well. The major exception here is the clustering techniques, but these techniques can pose a major advantage for models such as decision trees, which don't require the same level of pre-processing that parametric models might. Finally, if you find yourself suffering from high variance, you can always add more training data, with which your model is likely to get better.