Image Categorization Experiment

This project explores two methods of image categorization: prototype-based and examples-based categorization. The dataset used is a subset of the MNIST dataset containing only images labeled as 0 or 1. The project evaluates the accuracy of both methods by varying the percentage of training data used.

Project Structure

• main.py: The main script containing the functions and the experiment loop.
• train-images.idx3-ubyte: The file containing training images.
• train-labels.idx1-ubyte: The file containing labels for the training images.
• t10k-images.idx3-ubyte: The file containing test images.
• t10k-labels.idx1-ubyte: The file containing labels for the test images.
• Accuracy of prototype and examples-based categorization.png: The output plot showing the accuracy of both methods.

Additional Files

• all_digits_prototype_categorize.py: This script extends the prototype-based categorization to include all digit labels (0-9). It generates prototypes for each digit and calculates the categorization accuracy.
• all_digits_examples_categorize.py: This script extends the examples-based categorization to include all digit labels (0-9). It calculates the accuracy of categorization by comparing each test image to all training images.

Dependencies

To run the project, you need the following Python libraries:

• numpy
• matplotlib

Usage

Download and Place the Dataset Files

Ensure you have the following files in the same directory as the main.py script:

• train-images.idx3-ubyte
• train-labels.idx1-ubyte
• t10k-images.idx3-ubyte
• t10k-labels.idx1-ubyte

Run the Main Script

Execute the main script

Run the Additional Scripts

To run the scripts that include all digit labels (0-9):

all_digits_prototype_categorize.py

This script calculates the accuracy of prototype-based categorization for all digit labels (0-9) and displays the prototypes.

python all_digits_prototype_categorize.py

all_digits_examples_categorize.py

This script calculates the accuracy of examples-based categorization for all digit labels (0-9). Note that this script may take a long time to run due to the high computational complexity.

python all_digits_examples_categorize.py

View the Results

The scripts will generate plots showing the accuracy of the categorization methods. The main script will save the plot as Accuracy of prototype and examples-based categorization.png.

Functions

read_idx(filename)

Reads an IDX file and returns the data as a NumPy array.

create_dataset(train_images_path, train_labels, test_images, test_labels)

Reads the training and testing data from the specified files and returns the data, only including images with labels 0 or 1.

create_part_of_all_dataset(train_images_path, train_labels, test_images, test_labels, percentage)

Reads the training and testing data, and returns a subset of the training data based on the specified percentage, only including images with labels 0 or 1.

prototype_categorized(train_images, train_labels, test_images, test_labels)

Calculates the accuracy of prototype-based categorization by creating prototypes for each class and assigning the label of the closest prototype to the test images.

examples_categorize(train_images, train_labels, test_images, test_labels)

Calculates the accuracy of examples-based categorization by calculating the distance between each test image and each training image, and assigning the label of the closest training image to the test images.

Notes

• The examples_categorize function can take a long time to run due to the nested loops calculating distances between each test image and every training image.
• The script evaluates the accuracy of both methods by varying the percentage of training data used from 1% to 100%.

Results

The output plot shows the accuracy of both categorization methods. The x-axis represents the percentage of training data used, and the y-axis represents the accuracy.

The additional scripts provide a more comprehensive analysis by including all digit labels (0-9), giving a broader perspective on the performance of the categorization methods.