13. Perceptron class in sklearn
By Bernd Klein. Last modified: 16 Jun 2023.
In the previous chapter, we had implemented a simple Perceptron class using pure Python. The module
sklearn contains a
We saw that a perceptron is an algorithm to solve binary classifier problems. This means that a Perceptron is abinary classifier, which can decide whether or not an input belongs to one or the other class. E.g. "spam" or "ham". We accomplished this by linearly combining weights with the feature vector, i.e. the input.
It is amazing that the perceptron algorithm was already invented in the year 1958 by Frank Rosenblatt. The algorithm was implemented in custom-built hardware, called "Mark 1 perceptron". This hardware was designed for image recognition.
The invention has been extremely overestimated: In 1958 the New York Times wrote after a press conference with Rosenblatt: "New Navy Device Learns By Doing; Psychologist Shows Embryo of Computer Designed to Read and Grow Wiser"
What initially seemed very promising was quickly proved incapable of keeping its promises. Thes perceptrons could not be trained to recognise many classes of patterns.
Live Python training
Enjoying this page? We offer live Python training courses covering the content of this site.
Example: Perceptron Class
We will create with the help of
make_blobs a binary testset:
import matplotlib.pyplot as plt from sklearn.datasets import make_blobs n_samples = 500 data, labels = make_blobs(n_samples=n_samples, centers=([1.1, 3], [4.5, 6.9]), cluster_std=1.3, random_state=0) colours = ('green', 'orange') fig, ax = plt.subplots() for n_class in range(2): ax.scatter(data[labels==n_class][:, 0], data[labels==n_class][:, 1], c=colours[n_class], s=50, label=str(n_class))
We will split our testset into a learnset and testset:
from sklearn.model_selection import train_test_split datasets = train_test_split(data, labels, test_size=0.2) train_data, test_data, train_labels, test_labels = datasets
We will use not the
Perceptron class of
from sklearn.linear_model import Perceptron p = Perceptron(random_state=42) p.fit(train_data, train_labels)
We can calculate predictions on the learnset and testset and can evaluate the score:
from sklearn.metrics import accuracy_score predictions_train = p.predict(train_data) predictions_test = p.predict(test_data) train_score = accuracy_score(predictions_train, train_labels) print("score on train data: ", train_score) test_score = accuracy_score(predictions_test, test_labels) print("score on test data: ", test_score)
score on train data: 0.9125 score on test data: 0.96
Classifying the Iris Data with Perceptron Classifier
We want to apply the
Perceptron classifier on the iris dataset, which we had already used in our chapter on k-nearest neighbor
Loading the iris data set:
import numpy as np from sklearn.datasets import load_iris iris = load_iris()
array(['setosa', 'versicolor', 'virginica'], dtype='<U10')
We split the data into a learn and a testset:
from sklearn.model_selection import train_test_split datasets = train_test_split(iris.data, iris.target, test_size=0.2) train_data, test_data, train_labels, test_labels = datasets
Now, we create a Perceptron instance and fit the training data:
from sklearn.linear_model import Perceptron p = Perceptron(random_state=42, max_iter=30, tol=0.001) p.fit(train_data, train_labels)
Perceptron(max_iter=30, random_state=42)Perceptron(max_iter=30, random_state=42)
Now, we are ready for predictions and we will look at some randomly chosen random X values:
import random sample = random.sample(range(len(train_data)), 10) for i in sample: print(i, p.predict([train_data[i]]))
23  85  20  84  14  46  118  1  36  33 
from sklearn.metrics import classification_report print(classification_report(p.predict(train_data), train_labels))
precision recall f1-score support 0 1.00 0.48 0.65 75 1 0.05 1.00 0.09 2 2 1.00 0.93 0.96 43 accuracy 0.65 120 macro avg 0.68 0.80 0.57 120 weighted avg 0.98 0.65 0.75 120
from sklearn.metrics import classification_report print(classification_report(p.predict(test_data), test_labels))
precision recall f1-score support 0 1.00 0.74 0.85 19 1 0.17 1.00 0.29 1 2 1.00 1.00 1.00 10 accuracy 0.83 30 macro avg 0.72 0.91 0.71 30 weighted avg 0.97 0.83 0.88 30