This code base was developed as an independent project in collaboration with the CFA(Center for Astrophysics at Harvard) in order to detect promising transits in the kepler k2 time series data using CNNs. It is an ongoing project that shows promise. A potential paper in the Spring is to come, so stay tuned!
On Kaggle's transit dataset, it has reached a 98% accuracy. On real, unfiltered data (using an end-to-end methodology), it has reached around 75% accuracy on the binary classification task.