Lesson 6

• gini - how likely is it that you go into the sample and grab two items, they're the same.

• less processing for decision trees, no real dummy variables.

• don't have to care about outliers, categorical variables

• for tabular data, could start with decision trees for baseline approaches

• really hard to mess it up.

• can't split too deep because at some point the leaf nodes will not have much data.

• bagging: take the average of a number of models.

• find importances of the columns from random forests

• https://explained.ai/gradient-boosting/

1. Create an effective validation set
2. Iterate rapidly to find changes which improve results on a validation set.

dls.test_dl