what should be the order of class names in sklearn tree export function (Beginner question on python sklearn)

Question

I am trying a simple example with sklearn decision tree. I am giving "number,is_power2,is_even" as features and the class is "is_even" (of course this is stupid)

Here is the code

from sklearn import tree

features =[[1,0,0] , [2,1,1] , [3,0,0] , [4,1,1] , [5,0,0] , [6,0,1] ,  [900,0,1] , [1001,0,0]] #val,pow2,even
labels =  ['o'     ,  'e'    , 'o'     ,  'e'    ,  'o'    ,  'e'    ,   'e'      ,  'o'] #is even

clf = tree.DecisionTreeClassifier()
clf = clf.fit(features,labels)

print (clf.predict([[203,0,0]]))

import pydot

import pydotplus
from IPython.display import Image  
dot_data = tree.export_graphviz(clf, out_file=None, 
                     feature_names=['number','pow2','even'],  
                     class_names=['o','e'],  
                     filled=True, rounded=True,  
                     special_characters=True)  
graph = pydotplus.graph_from_dot_data(dot_data)  
# Image(graph.create_png())  
graph.write_pdf("1.pdf")

The decision tree correctly identifies even and odd numbers and the predictions are working properly.

The decision tree is basically like this (in pdf)

is_even<=0.5
        /
       /  
   label1  label2

The problem is this. The label1 is marked "o" and not "e". However if I put class_names in export function as

class_names=['e','o']

then, the result is correct. I thought the output should be independent of class_names order.

Am I doing something wrong, or does the class_names order matter. If the latter is true, what is the right order (for an arbitrary problem)

moomima · Accepted Answer

The names should be given in ascending numerical order.

Saw this in the code's docs:

class_names : list of strings, bool or None, optional (default=None)
    Names of each of the target classes in ascending numerical order.
    Only relevant for classification and not supported for multi-output.
    If ``True``, shows a symbolic representation of the class name.

Juan Carlos Ibañez · Answer

As described in the documentation. The names should be given in ascending order.

You can check the order used by the algorithm: the first box of the tree shows the counts for each class (of the target variable). The order es ascending of the class names

# class_names : list of strings, bool or None, optional (default=None)
# Names of each of the target classes in ascending numerical order.

class_names2pass = sorted(labels.unique())
print(class_names2pass)

Vlad Bezden · Answer

What you need to do is convert labels from string/char to numeric value. For instance 'o' = 0 and 'e' = 1

So your labels will look like this

labels =  [0, 1, 0, 1, 0, 1, 1, 0]

class_names should match those numbers in ascending numeric order

0 - 'o'
1 - 'e'

class_names=['o', 'e']

what should be the order of class names in sklearn tree export function (Beginner question on python sklearn)

3 Answers

Add your own answers!

Ask a Question