This is an implementation for our paper "Constrained Nonnegative Matrix Factorization based on Label Propagation for Data Representation".
Collected datasets can be loaded in Matlab.
Add the algorithms and tools into the path of Matlab and use the code in demo to accomplish relevant experiments.
You can also use the test results directly in file Saved Results which includes:
- comparison results with all state-of-the-art methods and ablation label propagation with ablate the label propagation (Data format:
number of cluster (9) * number of test(10)) - hyperparameters selection with label proportion, regularization and heat kernel (Data format:
number of cluster (9) * number of parameters(7 or 9) * number of test(10)) - relationships of label proportion with two hyperparameters label proportion with heat kernel and label proportion with regularization (Data format:
number of cluster (9) * number of label percent(9) * number of hyperparameters(7) * number of test(5)) - time consumption result (Data format:
1 * number of trails(10))