Code for the NASA ML challenge: Mars Spectrometry: Detect Evidence for Past Habitability
Two simple models included:
- MLP (only this model was used in the submission, no ensembles)
- CNN (very experimental, with the tabulard dataset converted to images using deep insight).
Achieved an AggLogLoss of 0.2170 during the first phase of the challenge with position 34/578 on the leaderboard.
Achieved an AggLogLoss of 0.0581 during the second phase of the challenge with position 24/677 on the leaderboard.