Based on Kaggle's competition from the famous MNIST data.
The file image-classification.ipynb is the main file for running the recognition.
So far, we have used Naive Bayes, which has performed pretty terribly (~.83). A few things that I would like to try:
- Simple SVM: https://www.kaggle.com/archaeocharlie/digit-recognizer/a-beginner-s-approach-to-classification
- Deep Neural Network (using Keros): https://www.kaggle.com/poonaml/deep-neural-network-keras-way
- Multi Layer Perceptron: https://www.kaggle.com/fchollet/simple-deep-mlp-with-keras/code/output
- Dimensionality Reduction (together used with SVM maybe) https://www.kaggle.com/arthurtok/interactive-intro-to-dimensionality-reduction
- Tensorflow: https://www.tensorflow.org/get_started/mnist/beginners