This project presents how to use Tensorflow to implement a standard NN for Kaggle mnist competition.
- config.py: consists of all settings, hyper-parameters
- data.py: to load data from csv and separate into data and label
- train.csv, test.csv: were downloaded from kaggle homepage, mnist project
- model.py: implement Model class to design expected model, get logit and weights or biases
- train.py: run this file to train and evaluate the model after setting, automatically save logs and checkpoint into result folder
- run.py: set the best step with the lowest cost and run this file to get prediction of test data and submit.
The best score of this model is 0.995 (Top 8%). Hyper-parameters tunning and more complex model would get better result.
Development enviroment:
- Window 10
- Anaconda 3
- Python 3.5
- Tensorflow 1.1
- Pycharm community