Tune Then Test
Choosing a setting and judging the final model must use different rows, or the judgment is too kind.
1. Split X and y into training and test parts with test_size=0.3 and random_state=0.
2. On the training part only, pick the max_depth from 1 to 5 of a decision tree (random_state=0) by 4-fold cross-validation.
3. Return the accuracy on the test part of the best tree, refitted on the whole training part.
Example 1
X [[0,0],[1,3],[2,6],[3,2],[4,5],[5,1],[6,4],[7,0],[0,3],[1,6],[2,2],[3,5],[4,1],[5,4],[6,0],[7,3],[0,6],[1,2],[2,5],[3,1],[4,4],[5,0],[6,3],[7,6],[0,2],[1,5],[2,1],[3,4],[4,0],[5,3],[6,6],[7,2],[0,5],[1,1],[2,4],[3,0],[4,3],[5,6],[6,2],[7,5]]
y [0,0,0,0,1,1,1,1,0,0,0,0,1,1,1,1,0,0,0,0,1,1,1,1,0,0,0,0,1,1,1,1,0,0,0,0,1,1,1,1]
returns 1.0
The label depends on one column crossing one value, so a single question is enough and the unseen rows are all classified correctly.
Constraints
40 <= len(X) <= 80- Labels are 0 and 1, about half each
Write it in your browser, against the clock. The model solution, stepped through line by line, comes after.
More scikit-learn problems
- Hold Some BackBeginner
- Learn Then PredictBeginner
- Names to NumbersBeginner
- Same ScaleBeginner
- Share RightBeginner
- Typical Squared MissBeginner
- What the Model LearnedBeginner
- Zero to OneBeginner