kaggle-plasticc

Code for the 14th place solution in the Kaggle PLAsTiCC competition.

See Modelling_approach.pdf for a detailed discussion of the modelling approach.

Quick-start guide to running the code

Total runtime is around 5.5 hours on a 24 Gb laptop.

• Download the code. Create a subfolder called data and save the csv files there.

• To reproduce the results exactly, create an environment with the specific package versions I used. (If you already have numpy, pandas, scikit-learn and lightgbm you can skip this step, but the results may differ slightly if you have different versions.) If you have conda, the easiest option is to build a conda environment using this command:

  conda env create environment.yml
  

  This will create an environment called plasticc-bt. The requirements.txt file is provided as well if you want to build an environment with pip.

• Run split_test.py to split the test data into 100 hdf5 files. They will be saved in an automatically created subfolder split_100 of the data folder. Takes around 15 minutes.

• Run calculate_features.py to calculate the features. This will generate 3 files in a folder called
  features (the folder is created automatically). Takes around 3.5 hours.

• Run predict.py to train the model and make predictions on the test set. Takes around 1.5 hours.

• Run scale.py to apply regularisation to the class 99 predictions and generate the final submission file. Takes a couple of minutes.