Commented Naive Bayes

Author: Kosaro

NaiveBayes.py

@@ -1,8 +1,47 @@
+# Naive Bayes Assignment
+# Oscar Kosar-Kosarewicz
+# opk18
+# 11/20/2020
+
+from sys import argv
 import re
 import numpy as np
 
 
+def main(train_path, test_path):
+    # train_path = 'data/NaiveBayes/breast_cancer.train.txt'
+    # test_path = 'data/NaiveBayes/breast_cancer.test.txt'
+
+    # train_path = 'data/NaiveBayes/led.train.txt'
+    # test_path = 'data/NaiveBayes/led.test.txt'
+
+    # scan files to get number of attributes and number of lines
+    train_attributes, train_lines = scan_file(train_path)
+    test_attributes, test_lines = scan_file(test_path)
+    num_attributes = max(train_attributes, test_attributes)
+
+    # read data from files
+    train_labels, train_x = read_data(train_path, num_attributes, train_lines)
+    test_labels, test_x = read_data(test_path, num_attributes, test_lines)
+
+    # Train model
+    class_weights = train(train_x, train_labels)
+
+    # predict labels
+    predicted_labels = predict(train_x, class_weights)
+    predicted_labels_test = predict(test_x, class_weights)
+
+    # print results
+    print_summary(train_labels, predicted_labels)
+    print_summary(test_labels, predicted_labels_test)
+
+
 def scan_file(filepath):
+    """
+    Scan the file to find the number of lines and attributes
+    :param filepath: input filepath
+    :return: number of attributes, number of lines
+    """
     max_index = 0
     num_lines = 0
     with open(filepath) as f:
@@ -14,6 +53,13 @@ def scan_file(filepath):
 
 
 def read_data(filepath, num_attributes, num_lines):
+    """
+    Read data from file
+    :param filepath: input file
+    :param num_attributes: number of attributes
+    :param num_lines: number of lines
+    :return: numpy ndarray with data
+    """
     labels = np.ndarray(num_lines, dtype=np.int)
     attributes = np.zeros((num_lines, num_attributes), dtype=np.int)
     with open(filepath) as f:
@@ -24,12 +70,64 @@ def read_data(filepath, num_attributes, num_lines):
     return labels, attributes
 
 
-train_path = 'data/NaiveBayes/breast_cancer.train.txt'
-test_path = 'data/NaiveBayes/breast_cancer.test.txt'
+def train(data, labels):
+    """
+    Trains a Naive Bayes classifier using Laplace smoothing.
+
+    :param data: training data
+    :param labels: training labels
+    :return: A 2 dimensional list containing dictionaries of probabilities for each
+    attribute for each class. shape is num_classes x num_attributes.
+    """
+    class_weights = []
+    # loop over classes/labels
+    for label in np.unique(labels):
+        label_count = np.sum(labels == label) + 2
+        attribute_weights = []
+        # loop over attributes
+        for i in range(data.shape[1]):
+            # get the unique values and respective counts for the attribute in rows with this label
+            values, counts = np.unique(data[labels == label, i], return_counts=True)
+            # Append dictionary with label probabilities for each value of this attribute
+            attribute_weights.append({value: (count + 1) / label_count for value, count in zip(values, counts)})
+        class_weights.append(attribute_weights)
+    return class_weights
+
+
+def predict(data, class_weights):
+    """
+    Use the class weights of the Naive Bayes classifier to predict labels.
+
+    :param data: data for prediction
+    :param class_weights: A 2 dimensional list containing dictionaries of probabilities for each
+    attribute for each class. shape is num_classes x num_attributes.
+    :return: predicted labels
+    """
+    probabilities = np.ndarray((data.shape[0], len(class_weights)))
+    for i, row in enumerate(data):
+        for r, label_weights in enumerate(class_weights):
+            probabilities[i, r] = np.product([label_weights[j].get(value, 0) for j, value in enumerate(row)])
+    result = np.argmax(probabilities, axis=1)
+    result[result == 0] = -1
+    return result
+
+
+def print_summary(true_y, predicted_y):
+    """
+    Print descriptive summary of Naive Bayes performance
+
+    :param true_y: True labels from dataset
+    :param predicted_y: predicted labels from model
+    :return:
+    """
+    matches = predicted_y[true_y == predicted_y]
+    differences = predicted_y[true_y != predicted_y]
+    true_positives = np.count_nonzero(matches[matches == 1])
+    true_negatives = np.count_nonzero(matches[matches == -1])
+    false_positives = np.count_nonzero(differences[differences == 1])
+    false_negatives = np.count_nonzero(differences[differences == -1])
+    print(f'{true_positives} {false_negatives} {false_positives} {true_negatives}')
 
-train_attributes, train_lines = scan_file(train_path)
-test_attributes, test_lines = scan_file(test_path)
-num_attributes = max(train_attributes, test_attributes)
 
-labels_train, attributes_train = read_data(train_path, num_attributes, train_lines)
-labels_test, attributes_test = read_data(test_path, num_attributes, test_lines)
+if __name__ == '__main__':
+    main(argv[1], argv[2])