ytsvetko/classify

python scikit-learn wrapper with json-type input features

A wrapper to python scikit-learn classifiers (e.g., SVM, Logistic Regression, Random Forest, Gradient Boosted Regression Trees) that works with sparse input features in json format

Input files format:

train_features file in json format:

sample1 \t {"feat0": 0.1, "feat1": -0.1}
sample2 \t {"feat0": 5, "feat1": 0.5, "feat2": -0.2}

train_labels file:

sample1 \t LABEL
sample2 \t LABEL

Usage:

1. 10-fold cross validation

   ./classify.py --classifier RandomForest
   --train_features data/train.feat
   --train_labels data/train.labels
   --num_cross_validation_folds 10

2. Training a model and saving it

   ./classify.py --classifier SVM
   --train_features data/train.feat
   --train_labels data/train.labels
   --dump_classifier_filename out/svm.model

3. Prediction using a trained model

   ./classify.py --load_classifier_filename out/svm.model
   --test_features data/test.feat
   --test_predicted_labels_out out/test.labels.out
   --write_posterior_probabilities

4. Evaluate predictions: Accuracy & confusion matrix

   ./classify.py --load_classifier_filename out/svm.model
   --test_features data/test.feat
   --golden_labels data/test.labels