Constructing a confusion matrix from data without sklearn

I am trying to construct a confusion matrix without using the sklearn library. I am having trouble correctly forming the confusion matrix. Here's my code:

def comp_confmat(): currentDataClass = [1,3,3,2,5,5,3,2,1,4,3,2,1,1,2] predictedClass = [1,2,3,4,2,3,3,2,1,2,3,1,5,1,1] cm = [] classes = int(max(currentDataClass) - min(currentDataClass)) + 1 #find number of classes for c1 in range(1,classes+1):#for every true class counts = [] for c2 in range(1,classes+1):#for every predicted class count = 0 for p in range(len(currentDataClass)): if currentDataClass[p] == predictedClass[p]: count += 1 counts.append(count) cm.append(counts) print(np.reshape(cm,(classes,classes))) 

However this returns:

[[7 7 7 7 7] [7 7 7 7 7] [7 7 7 7 7] [7 7 7 7 7] [7 7 7 7 7]] 

But I don't understand why each iteration results in 7 when I am reseting the count each time and it's looping through different values?

This is what I should be getting (using the sklearn's confusion_matrix function):

[[3 0 0 0 1] [2 1 0 1 0] [0 1 3 0 0] [0 1 0 0 0] [0 1 1 0 0]] 

3 Answers

You can derive the confusion matrix by counting the number of instances in each combination of actual and predicted classes as follows:

import numpy as np def comp_confmat(actual, predicted): # extract the different classes classes = np.unique(actual) # initialize the confusion matrix confmat = np.zeros((len(classes), len(classes))) # loop across the different combinations of actual / predicted classes for i in range(len(classes)): for j in range(len(classes)): # count the number of instances in each combination of actual / predicted classes confmat[i, j] = np.sum((actual == classes[i]) & (predicted == classes[j])) return confmat # sample data actual = [1, 3, 3, 2, 5, 5, 3, 2, 1, 4, 3, 2, 1, 1, 2] predicted = [1, 2, 3, 4, 2, 3, 3, 2, 1, 2, 3, 1, 5, 1, 1] # confusion matrix print(comp_confmat(actual, predicted)) # [[3. 0. 0. 0. 1.] # [2. 1. 0. 1. 0.] # [0. 1. 3. 0. 0.] # [0. 1. 0. 0. 0.] # [0. 1. 1. 0. 0.]] 

In your innermost loop, there should be a case distinction: Currently this loop counts agreement, but you only want that if actually c1 == c2.

Here's another way, using nested list comprehensions:

currentDataClass = [1,3,3,2,5,5,3,2,1,4,3,2,1,1,2] predictedClass = [1,2,3,4,2,3,3,2,1,2,3,1,5,1,1] classes = int(max(currentDataClass) - min(currentDataClass)) + 1 #find number of classes counts = [[sum([(currentDataClass[i] == true_class) and (predictedClass[i] == pred_class) for i in range(len(currentDataClass))]) for pred_class in range(1, classes + 1)] for true_class in range(1, classes + 1)] counts 

[[3, 0, 0, 0, 1], [2, 1, 0, 1, 0], [0, 1, 3, 0, 0], [0, 1, 0, 0, 0], [0, 1, 1, 0, 0]] 

Here is my solution using numpy and pandas:

import numpy as np import pandas as pd true_classes = [1, 3, 3, 2, 5, 5, 3, 2, 1, 4, 3, 2, 1, 1, 2] predicted_classes = [1, 2, 3, 4, 2, 3, 3, 2, 1, 2, 3, 1, 5, 1, 1] classes = set(true_classes) number_of_classes = len(classes) conf_matrix = pd.DataFrame( np.zeros((number_of_classes, number_of_classes),dtype=int), index=classes, columns=classes) for true_label, prediction in zip(true_classes ,predicted_classes): # Each pair of (true_label, prediction) is a position in the confusion matrix (row, column) # Basically here we are counting how many times we have each pair. # The counting will be placed at the matrix index (true_label/row, prediction/column) conf_matrix.loc[true_label, prediction] += 1 print(conf_matrix.values) 

[[3 0 0 0 1] [2 1 0 1 0] [0 1 3 0 0] [0 1 0 0 0] [0 1 1 0 0]]