This is a project for the Kaggle competition Don't Get Kicked!.

The project uses:

• Python (tested with 2.6 and 2.7)
• numpy (tested with 1.6.1, does not work with 1.5.1)
• FANN (2.1.0)
• argparse (tested with 1.1)
• matplotlib (tested with 1.0.1)
  • only needed for: zipcodes.py, states.py

The project currently consists of the following files:

Main files

• load.py Loads the training data and transforms it into training data for FANN. (Step 1)
• train.py
  • Trainer. Runs FANN on generated training data and generates neural network. (Step 2)
  • Tester. Runs generated neural network on generated testing data. (Step 3)

Note: Training data and testing data are basically the same (they have the same format), but it is recommended to have them in seperate files so you don't test on the same data you trained on.

Auxiliary files

• zipcodes.py -- Plots the IsBadBuy score for each zipcode.
• states.py -- Plots the IsBadBuy score for each state.

1. Make training file (reads from training.csv outputs to data/train.dat)

    python load.py
   

2. Make test file (reads from training.csv outputs to data/train.dat)

    python load.py --test-data --range 2001,4000
   

   • The training data is by default made from line 1-2000 in training.csv, here we choose line 2001-4000

3. Train the neural network (reads from data/train.dat and outputs the neural network to networks/default.net)

    python train.py
   

4. Test the neural network (reads from data/test.dat using the neural network in networks/default.net)

    python train.py --test
   

1. Make training file (reads lines 5000-6000 from training.csv outputs to data/minimal_train.dat)

   The training file's input neurons is specified in config/datamapping/minimal.py

    python load.py --data-mapping minimal --range=5000,6000 --train-file 'data/minimal_train.dat'
   

2. Make test file (reads lines 3700-4700 from training.csv outputs to data/minimal_train.dat)

   The test file's input neurons is specified in config/datamapping/minimal.py

    python load.py --data-mapping minimal --range=3700,4700 --test-file 'data/minimal_test.dat'
   

3. Train the neural network (reads from data/minimal_train.dat and outputs the neural network to networks/minimal.net)

   The neural network's learning rate, hidden layer etc. is read from config/learn/minimal.py

    python train.py --train-file 'data/minimal_train.dat' --neural-config minimal --network-file 'networks/minimal.net'
   

4. Test the neural network (reads from data/minimal_test.dat using the neural network in networks/minimal.net) Also gives line-by-line output of the test results

    python train.py --test-file 'data/minimal_test.dat' --network-file 'networks/minimal.net' --full
   

The main file. Loads the training data and transforms it into training data for FANN.

Move along -- nothing to see here...

Move along -- nothing to see here...