cgearhart/kaggle-digits

Kaggle MNIST handwritten digit recognition project

setup

1. Create virtual environment

$ virtualenv env
$ . env/bin/activate

2. Install required libraries

(env)$ pip install theano
(env)$ git clone git://github.com/lisa-lab/pylearn2.git
(env)$ cd pylearn2
(env)$ python setup.py develop

3. Modify environment variables

   • add the following lines to env/bin/activate -- they're required for python to find the random number generator in the configuration yaml file

PYTHONPATH=""
export PYTHONPATH

4. Generate the data files

   • Download the data (from kaggle); the scripts expect the data files to be in the "./data" directory relative to the .yaml and make_pylearn_data.py files
   • Run the data generator

(env)$ python make_pylearn_data.py

5. (optional) Install & configure CUDA

   • download the installer from https://developer.nvidia.com/cuda-downloads
   • add required envronmental variables http://docs.nvidia.com/cuda/cuda-getting-started-guide-for-mac-os-x/index.html#installation
   • set up ~/.theanorc

[global]
device=gpu0
force_device=False
openmp=False
floatX=float32

6. Run the model trainer

local:

(env)$ pylearn2-train conv2.yaml

ssh:

(env)$ nohup caffeinate pylearn2-train conv2.yaml &

7. Predict new data -- see pylearn2/scripts/mlp/predict_csv.py