Digital data-set generated using Python PIL and OS fonts for AI semester project.
test-data-519/: This folder contains all the dataset images in their respective folders.
TESTSet.py: This code is used to generate the dataset. This takes all fonts from the operating system and writes them to the 10x10 image to generate the dataset image.
MLP.py: This code contains the code to load the dataset (in Python) and then test that dataset in Keras and Scikit-learn.
This result were collected using MLP.py, goal of this result were to match Scikit-learn and Keras result under same architecture. While with default parameter and same architecture, result difference is high (Scikit-learn is better).
Number of samples in training set: 4150, number of samples in test set: 1040
('Train Accuracy:', 0.98168674698795177)
('Test Accuracy:', 0.85673076923076918)
('Layers:', 3)
('Output Layer size: ', 10)
('Number of Iteration: ', 200)
('Output Activation: ', 'softmax')
Number of samples in training set: 4150, number of samples in test set: 1040
('Train accuracy:', 0.98168674713157744)
('Test accuracy:', 0.85961538461538467)
('Layers:', 3)
('Output Layer size: ', 10)
('Number of Epochs: ', 200)
('Output Activation: ', 'softmax')
This project is licensed under the MIT License - see the LICENSE file for details