This repository contains source code used for the "Ensemble Wrapper Subsampling for Deep Modulation Classification" project
- radioml_ews: Ensemble Wrapper Subsampling
- radioml_fisher: Subsampling using the Fisher Score
- radioml_fqi: Subsampling using the Feature Quality Index (FQI)
- radioml_laplacian: Sampling using the Laplacian Score
- radioml_rfs: Subsampling using Efficient and Robust Feature Selection (RFS)
Download the RML2016.10b_dict.dat zip file (containing the original dataset) from www.kaggle.com/sramjee/rml10b and extract it in the data directory.
Ranker Models used in the Subsampler Nets used in Ensemble Wrapper Subsampling.
Subsampler Nets used in Ensemble Wrapper Subsampling. These subsampler nets use the Ranker Models in the models directory, perform the subsampling, and store data in the ranker_samples directory.
Ensemble Wrapper Subsampling samples selected using the CNN, CLDNN, and ResNet Subsampler Nets with 1/2, 1/4, 1/8, 1/16, and 1/32 subsampling rates. Download the ranker_samples zip file (containing the subsampled data) from https://www.kaggle.com/sramjee/ranker-samples and extract it in the root (i.e. dds) directory.