Use RandomForestClassifier to understand the output of TreeInterpreter
I applied a random forest classifier to get features that contribute to specific rows in the dataset. However, I get 2 eigenvalues instead of one. I’m not quite sure why. Here is my code.
import numpy as np
import pandas as pd
from sklearn.datasets import make_classification
from sklearn.ensemble import RandomForestClassifier
from treeinterpreter import treeinterpreter as ti
from treeinterpreter import treeinterpreter as ti
X, y = make_classification(n_samples=1000,
n_features=6,
n_informative=3,
n_classes=2,
random_state=0,
shuffle=False)
# Creating a dataFrame
df = pd. DataFrame({'Feature 1':X[:,0],
'Feature 2':X[:,1],
'Feature 3':X[:,2],
'Feature 4':X[:,3],
'Feature 5':X[:,4],
'Feature 6':X[:,5],
'Class':y})
y_train = df['Class']
X_train = df.drop('Class',axis = 1)
rf = RandomForestClassifier(n_estimators=50,
random_state=0)
rf.fit(X_train, y_train)
print ("-"*20)
importances = rf.feature_importances_
indices = X_train.columns
instances = X_train.loc[[60]]
print(rf.predict(instances))
print ("-"*20)
prediction, biases, contributions = ti.predict(rf, instances)
for i in range(len(instances)):
print ("Instance", i)
print ("-"*20)
print ("Bias (trainset mean)", biases[i])
print ("-"*20)
print ("Feature contributions:")
print ("-"*20)
for c, feature in sorted(zip(contributions[i],
indices),
key=lambda x: ~abs(x[0].any())):
print (feature, np.round(c, 3))
print ("-"*20)
This is the output of my code. Can someone explain why bias and features output 2 values instead of one?
--------------------
[0]
--------------------
Instance 0
--------------------
Bias (trainset mean) [ 0.49854 0.50146]
--------------------
Feature contributions:
--------------------
Feature 1 [ 0.16 -0.16]
Feature 2 [-0.024 0.024]
Feature 3 [-0.154 0.154]
Feature 4 [ 0.172 -0.172]
Feature 5 [ 0.029 -0.029]
Feature 6 [ 0.019 -0.019]
Solution
Since you have a Class 2 classification problem, you will get an array of biases and feature contributions of length 2.
As clearly explained in this blog post by the package creator, in 3 of the iris dataset In the case of classes, you get arrays of length 3 (i.e. one array element per class):
from treeinterpreter import treeinterpreter as ti
import numpy as np
from sklearn.ensemble import RandomForestClassifier
from sklearn.datasets import load_iris
iris = load_iris()
rf = RandomForestClassifier(max_depth = 4)
idx = range(len(iris.target))
np.random.shuffle(idx)
rf.fit(iris.data[idx][:100], iris.target[idx][:100])
prediction, bias, contributions = ti.predict(rf, instance)
print "Prediction", prediction
print "Bias (trainset prior)", bias
print "Feature contributions:"
for c, feature in zip(contributions[0],
iris.feature_names):
print feature, c
Give:
Prediction [[ 0. 0.9 0.1]]
Bias (trainset prior) [[ 0.36 0.262 0.378]]
Feature contributions:
sepal length (cm) [-0.1228614 0.07971035 0.04315104]
sepal width (cm) [ 0. -0.01352012 0.01352012]
petal length (cm) [-0.11716058 0.24709886 -0.12993828]
petal width (cm) [-0.11997802 0.32471091 -0.20473289]
Formula
prediction = bias + feature_1_contribution + ... + feature_n_contribution
From TreeInterpreter applies to each class, in the case of classification problems; Therefore, for a class k classification problem, the length of the corresponding array is k (in your example, k=2, and for the IRIS dataset, k=3).