Learning Decision Tree in scikit-learn Package

Today, I spent more time on how to specify and visualize decision tree classifier in scikit-learning package and finally have a better understanding. With some tweaking, sklearn.tree module works pretty well with pandas package that I am actively learning. Below is a piece of revised code that is close to what we could use in real-world problems.

In [1]: # LOAD PACKAGES

In [2]: from sklearn import tree

In [3]: from pandas import read_table, DataFrame

In [4]: from os import system

In [5]: # IMPORT DATA

In [6]: data = read_table('/home/liuwensui/Documents/data/credit_count.txt', sep = ',')

In [7]: # DEFINE THE RESPONSE

In [8]: Y = data[data.CARDHLDR == 1].BAD

In [9]: # DEFINE PREDICTORS

In [10]: X = data.ix[data.CARDHLDR == 1, "AGE":"EXP_INC"]

In [11]: # SPECIFY TREE CLASSIFIER

In [12]: dtree = tree.DecisionTreeClassifier(criterion = "entropy", min_samples_leaf = 500, compute_importances = True)

In [13]: dtree = dtree.fit(X, Y)

In [14]: # PRINT OUT VARIABLE IMPORTANCE

In [15]: print DataFrame(dtree.feature_importances_, columns = ["Imp"], index = X.columns).sort(['Imp'], ascending = False)
               Imp
INCOME    0.509823
INCPER    0.174509
AGE       0.099996
EXP_INC   0.086134
ACADMOS   0.070118
MINORDRG  0.059420
ADEPCNT   0.000000
MAJORDRG  0.000000
OWNRENT   0.000000
SELFEMPL  0.000000

In [16]: # OUTPUT DOT LANGUAGE SCRIPT

In [17]: dotfile = open("/home/liuwensui/Documents/code/dtree2.dot", 'w')

In [18]: dotfile = tree.export_graphviz(dtree, out_file = dotfile, feature_names = X.columns)

In [19]: dotfile.close()

In [20]: # CALL SYSTEM TO DRAW THE GRAPH

In [21]: system("dot -Tpng /home/liuwensui/Documents/code/dtree2.dot -o /home/liuwensui/Documents/code/dtree2.png")
Out[21]: 0

dtree2