What does model calibration mean?
What does model calibration mean?
Recipe Objective
Generally, for any classification problem, we predict the class value that has the highest probability of being the true class label. However, sometimes, we want to predict the probabilities of a data instance belonging to each class label. This type of problems can easily be handled by calibration curve. It support models with 0 and 1 value only.
So this recipe is a short example on what does caliberation mean. Let's get started.
Step 1 - Import the library
from sklearn import datasets
from sklearn.model_selection import train_test_split
from sklearn.datasets import load_breast_cancer
from sklearn.tree import DecisionTreeClassifier
from sklearn.calibration import calibration_curve
import matplotlib.pyplot as plt
Let's pause and look at these imports. We have exported train_test_split which helps in randomly breaking the datset in two parts. Here sklearn.dataset is used to import one classification based model dataset. Also, we have exported calibration_curve to calibrate our model.
Step 2 - Setup the Data
X,y=load_iris(return_X_y=True)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.25)
Here, we have used load_iris function to import our dataset in two list form (X and y) and therefore kept return_X_y to be True. Further with have broken down the dataset into 2 parts, train and test with ratio 3:4.
Now our dataset is ready.
Step 3 - Building the model
model =DecisionTreeClassifier(criterion ='entropy', max_features = 2)
We have simply built a classification model with =DecisionTreeClassifier with criterion as entropy and max_feature to be 2.
Step 4 - Fit the model and predict for test set
model.fit(X_train, y_train)
y_pred= model.predict(X_test)
Here we have simply fit used fit function to fit our model on X_train and y_train. Now, we are predicting the values of X_test using our built model.
Step 5 - Calibrating our model
x, y = calibration_curve(y_test, y_pred, n_bins = 10, normalize = True)
Now we are calibrating our predicted value to the actual value. n_bins refers for number of bins to discretize the [0, 1] interval. Also, we are normalizing y_pred in the [0,1] interval.
Step 5 - Plotting results
plt.plot([0, 1], [0, 1], linestyle = '--', label = 'Ideally Calibrated')
plt.plot(y, x, marker = '.', label = 'Decision Tree Classifier')
plt.xlabel('Average Predicted Probability in each bin')
plt.ylabel('Ratio of positives')
plt.legend()
plt.show()
Here, first we have plotted the Ideally calibrated curve which will a straight line between 0 and 1. Now, we plot our calibrated curve of this particular model. he x-axis represents the average predicted probability in each bin. The y-axis is the ratio of positives (the proportion of positive predictions).
Step 7 - Lets look at our dataset now
Once we run the above code snippet, we will see:
Scroll down to the ipython file to visualize the results.
Clearly, the model built is highly efficient on any unknown set.
