MNIST Sample Models

This repo contains (a) quick implementation(s) of digit recognition from the MNIST dataset. This is repo is a more accurate representation of my coding ability after my internship at Pinterest :)

Model Plans

1. embedding based basic model (try something different)
   a. Set up tensorboard logging

2. Auto-encoder (most guides of this don't show actual implementation. Will implement based on the basic model as multi-task learning)

3. Mixture of Softmaxes (This shouldn't help but just for fun I'll implement)
   (https://arxiv.org/pdf/1711.03953.pdf)

4. CNN (Very typical to be used for this problem)

Overview

This repo contains a few models used to classify hand-written digits from the MNIST dataset as well as their performance metrics.

Required Packages

Python 3.6.3

tensorflow==1.10.0
tensorboard==1.10.0
numpy==1.14.5

Running

python3 path/to/desired/model.py

Future Plans

1. Saving checkpoints so that models can be trained and retrained