Finish report for project 3.

README.md

@@ -34,7 +34,8 @@ unzip -n zips/*.zip -d ./
 # Reports
 
 - [Report 1](./martingale/martingale.md)
-- [Report 2](./optimize_something/readme.md)
-- [Report 3](#)
+- [Report 2](./optimize_something/optimize_something.md)
+- [Report 3](./assess_learners/assess_learners.md)
+- [Report 4](#)
 
 

assess_learners/assess_learners.md

@@ -0,0 +1,44 @@
+# Report
+
+## Experiment 1
+
+Significant overfitting occurs for leaf sizes smaller than five. The chart shows
+that the root-mean-square-error is significantly higher for the test data
+(`rmse_out`) for leaf sizes smaller than five.
+
+Between five and nine, the error for the test data is only slightly higher, so
+there is small overfitting. Beyond that, the errors increase, and the error for
+the test data is lower than for the training data. In other words, there is no
+more overfitting for leaf sizes greater than nine.
+
+![](figure_1.png)
+
+
+## Experiment 2
+
+For all bag sizes, the difference of the RMSE for the training data and the test
+data is smaller than without bagging. The test data still has a lower RMSE up to
+a leaf size of five. For greater leaf sizes, the RMSE for the test data is
+smaller than for the training data for all bag sizes, so there is no
+overfitting.
+
+![](figure_2.png)
+
+![](figure_3.png)
+
+![](figure_4.png)
+
+![](figure_5.png)
+
+
+## Experiment 3
+
+The Random Tree learner has a correlation of one for the training data. In other
+words, it fits the training data perfectly. Consequently, the correlation for
+the test data is worse than for the Decision Tree learner. The DT learner has a
+higher correlation than the RT for all other leaf sizes, both for the training
+and the test data.
+
+![](figure_6.png)
+
+![](figure_7.png)