8000 GitHub - Mr-Yuppie/SPBL: The matlab code for the IJCAI-16 paper "Self-Paced Boost Learning for Classification"
[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
Skip to content
/ SPBL Public

The matlab code for the IJCAI-16 paper "Self-Paced Boost Learning for Classification"

Notifications You must be signed in to change notification settings

Mr-Yuppie/SPBL

Repository files navigation

SPBL

The matlab code for the IJCAI-16 paper "Self-Paced Boost Learning for Classification"

  1. Store the data as "Data/fea.mat" (in the "Data/" directory), which has two matrices: fea: n-by-d matrix, where each row is the feature of a sample; gnd: n-by-1 vector, where each element is the class label index of a sample.

  2. Run "Gen_Split.m" to generate the split of traning/validation/test set of the data. Set the "train_ratio" and "vali_ratio" variables for the proportions of the training and the validation samples, respectively.

  3. (Optional) Run "Gen_Noise" to generate label noise in the training set. Set the "n_ratio" variable for the proportion of the noisily labeled samples.

  4. Run "SPBLmain.m" to train the SPBL model with the training set and test the learned model on the test set. The output experimental results: ResObj.test_err: test error rate at each iteration, with the first (second) column as the top-1 (top-5) error rates. ResObj.vali_err: validation error rate at each iteration, with the first (second) column as the top-1 (top-5) error rates. OptRes: struct varaible. The optimal results on the test set based on the validation.

About

The matlab code for the IJCAI-16 paper "Self-Paced Boost Learning for Classification"

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published
0