/mnist-exploration

A short exploration of MNIST dataset with PyTorch

Primary LanguageTeX

Dependencies

With Conda, simply run

conda install pytorch torchvision -c soumith
conda install pandas seaborn #For plotting
conda install scikit-learn #For computing confusion matrix

A more detailed installation guide of PyTorch for different environments can be found in PyTorch. Experiment codes are tested in Ubuntu 16.04 and plotting scripts are run in Mac OS X environment, with Python 3.5/ 3.6.

Code Structures

  • net.py, utils.py - the builiding blocks of the experiments, mainly the network structures
  • exp_.py - the experiments codes for the three sections
  • exp_plot.py - the plotting scripts for each section
  • out/ - the intermediate folder storing data for plots and tables in the report
  • report/ - source files for the report

Running the Codes

Simpy invoke

python exp_.py

to reproduce the results.