- original image size (1 x 10 x 10), it is symmetric, use only the right-top quarter
- 0: success, 1: failure (based on the top 20 quantile value of bandgap)
- imbalanced data (80/20)
- CNN with data augmentation (i.e., random flip and random rotation) and imbalanced data sampler(pytorch, TLClassifier.py)
- Bayes CNN: Bayes by Backprop, local reparameterization trick, variational dropout
- SVM, KNN, Random Forest, and MLP with hyperparameter tuning (sklearn, sklearnClassifier.py)
- CNN regressor: zero_padding enlarging mode in Resize(), target variable normalization techniques (sklearn)