22 lines
1.1 KiB
Org Mode
22 lines
1.1 KiB
Org Mode
|
Modify the given Jupyter notebook on decision trees on Iris data and
|
||
|
|
perform the following tasks:
|
||
|
|
|
||
|
|
1. get an artificial inflation of some class in the training set by
|
||
|
|
a given factor: 10 (weigh more the classes virginica e versicolor
|
||
|
|
which are more difficult to discriminate). Learn the tree in these
|
||
|
|
conditions.
|
||
|
|
2. modify the weight of some classes (set to 10 the weights for
|
||
|
|
misclassification between virginica into versicolor and vice versa)
|
||
|
|
and learn the tree in these conditions. You should obtain similar
|
||
|
|
results as for step 1.
|
||
|
|
3. learn trees but avoid overfitting (by improving the error on the
|
||
|
|
test set) tuning the parameters on: the minimum number of samples
|
||
|
|
per leaf, max depth of the tree, min_impurity_decrease parameters,
|
||
|
|
max leaf nodes, etc.
|
||
|
|
4. build the confusion matrix of the created tree models on the
|
||
|
|
test set and show them.
|
||
|
|
5. build the ROC curves (or coverage curves in coverage space) and
|
||
|
|
plot them for each tree model you have created: for each model you
|
||
|
|
have to build three curves, one for each class, considered in turn
|
||
|
|
as the positive class.
|