Building The Model

PLAsTiCC is a timeseries dataset. Here's the process to create a model using timeseries data:

  • Preprocessing cleans up the data and establishes base attributes.
  • Feature Engineering is the task of creating features. Thousands of features (mostly hand-crafted) are common in PLAsTiCC submissions.
  • Feature Engineering also eliminates the time domain, replacng it with reduction features (think of "period" replacing a series of amplitudes).
  • A metric measures the model's performance. Metrics need to be established up front!

    Google's Tensorflow provides the framework.

    ???

    references:
  • https://www.tensorflow.org/
  • https://github.com/cwinsor/kaggle_plasticc.git