Congratulations on your choice of a interesting course if you find this repository. Don't be tired of his f**g lectures, or you would be failed in the final exam...
- features:
output/data/dataset_train.npy dataset_test.npy (3558 * num_of_features) - labels:
output/data/target_train.npy dataset_test.npy (3558 * 1)
using 1 fc-layer.
using 1 fc-layer and hinge loss.
using several fc-layers and crossentropy loss.
- LR: 88% (with l2 norm) 87% (without norm)
- SVM: 89%
- NN: 91% (with l2 norm, dropout(0.5), relu activation funciton)
Results could be improved with some more parameters adjustment.