mrecos / plankton_ndsb Goto Github PK

Created by Jeff Hebert 12/30/14

For Kaggle Competition: http://www.kaggle.com/c/datasciencebowl

This is basic starter code to create a plankton classification model based on image dimensions and image density. This code should produce a results file with public score of 3.397518. Surely you can come up with a better model!

1. Download and unzip the competition data
2. Set the data_dir variable to the path of the data directory
3. Run the code. It will take 30-50 minutes, limited by IO of the thousands of tiny files
4. Customize the extract_stats function to add your custom variables
5. Remember to add new variables to train_data and test_data

1. Use cross validation to get a better model.
2. Please let me know if you find this starter code helpful.
3. Remember to use Rprof() and microbenchmark() when your code slows down.
4. Remember to use some form of version control and rename your submission files.
5. Please let me know if you find some methods to speed up the code.