GitHub - moeabdol/cnn-mnist: Python implementation of a Convolutional Neural Network running on the MNIST dataset.

moeabdol / cnn-mnist Public

Notifications You must be signed in to change notification settings
Fork 5
Star 6

Python implementation of a Convolutional Neural Network running on the MNIST dataset.

Notifications

Name		Name	Last commit message	Last commit date
Latest commit History 1 Commit
CNN.py		CNN.py
Logistic_Regression.py		Logistic_Regression.py
MLP.py		MLP.py
README		README
Utils.py		Utils.py
main.py		main.py
mnist.pkl.gz		mnist.pkl.gz

Repository files navigation

This code implements a full Convolutional Neural Network with a Multi-Layer
Perceptron and a Logistic Regression classifier. The model trains on the famous
MNIST dataset.

To run the code it is better to use iPython.

To create the CNN model, run ipython from within the project directory then:
from main import *

Once the model is built and ipython returns the prompt type in the follwoing to train the network:
train()

If you want to test how well the classifier classifies the test set type in:
test()

To actually see some misclassified examples, make sure you train the network first and then type in:
show_misclassified()

Note:
If you want to run the code directly using python2.7, then you have to slightly modify the code in main.py.
I happen to use the interactive iPython because it is easier for me to debug. Most of the code in this 
project is readily available at www.deeplearning.net. I modified the code so I can show missclassified 
examples. Enjoy!