Yushi Wang, Jie Fu, Yumeng Yin

This competition was mainly done by Yushi Wang.

Due to the configuration problems with neon, we haven't had enough time to tune the hyperparameters. Our solution is first do some filtering in the frequency-domain, and then train a CNN in the space-domain.

We put code, experimental procedure, documents, discussion and resources here. As for project management, we use Trello. For light chat, we stick with Gitter.

• SKlearn
• Neon, https://github.com/NervanaSystems/neon
• Theano, http://deeplearning.net/software/theano/
• Keras
• RoBO
• Hyperopt, https://github.com/hyperopt/hyperopt

1. https://www.kaggle.com/anlthms/grasp-and-lift-eeg-detection/convnet-0-89 Sample code with Neon

1. Do filtering in the frequency-domain
2. Transform back to space-domain, and train deep neural networks
3. Tune hyperparameters using RoBO or hyperopt

When submitting, we create our own folders, such as fj_submit - rnn_baseline.

Dense documentation is desired. So put all the thoughts and related stuff in markdown files associated with the code and data.