cnn-mnist

Instalation

git clone https://github.com/kapiw04/cnn-mnist.git
pip install -r requirements.txt

Usage

To use default hyperparameters for CNN, which I found to be the best ones for the highest accuracy run this command

python3 cnn.py

To configure a parameter add <paremeter name>=<value>

Parameter	Description	Default
batch_size	Number of samples per gradient update	64
learning_rate	Learning rate for the optimizer	0.01
num_epochs	Number of epochs to train the model	10
momentum	Momentum for the optimizer	0.9
kernel_size	The size of the convolutional kernel	3
dropout_rate	Dropout rate for regularization	0.5
l1_size	Size of the first linear layer	128
l2_size	Size of the second linear layer	64
conv_1_size	Size of the first convolutional layer	32
conv_2_size	Size of the second convolutional layer	64

Example:

python3 cnn.py batch_size=32 dropout_rate=0.4

Preview of the model

python3 gui.py