INTRO

data mining project 2 implemented with SGD and SVM kernel implemented with sklearn class RBFSampler,Nystroem,AdditiveChi2Sampler,SkewedChi2Sampler Final version with AdditiveChi2Sampler

HOW TO RUN

1. download repository to your local drive and cd to the folder data_proj2.
2. Copy and paste you previous data files to the folder: training_set.txt test_data.txt test_label.txt
3. To compute weights: cat training_set.txt | python mapper.py | python reducer > weights.txt
4. As an exmaple, the computed weights is already in the folder: weights_4.txt
5. To compute accuracy: evaluate.py weights_4.txt test_data.txt test_label.txt

RESULT

• The current method AdditiveChi2Sampler have accuracy = 0.80463
• SkewedChi2Sampler(skewedness=1.0, n_components=100, random_state=1) accuracy = 0.703116
• tune gamma and n_components of RBFSampler to get better accuracy
  • when RBFSampler(gamma=0.1, n_components=100), accuracy = 0.687130
  • when RBFSampler(gamma=0.1, n_components=200), accuracy = 0.681268
  • when RBFSampler(gamma=1, n_components= 100), accuracy = 0.714028
• without any kernel transform accuracy = 0.74747