SLeDGe

SLeDGe (Semi-supervised Learning on Data stream with Graph structure learning) is a framework designed for semi-supervised learning on data streams. It simultaneously learns a predictive model and an adaptive graph structure, specifically optimized for scenarios with scarce labels and limited storage capacity.

• Label Ratio: Supported ratios are 0.01 or 0.001.
• Temporal Weighting: Append the -time flag to enable temporal weighting.

• Using the best configuration:

  python src/main_SLeDGe.py --dataset={dataset_name} --gpu={gpu_id} --label-ratio={label_ratio}

• Running all configurations:

  python src/main_SLeDGe_full.py --dataset={dataset_name} --gpu={gpu_id} --label-ratio={label_ratio}

• Using the best configuration:

  python src/main_SLeDGe_Light.py --dataset={dataset_name} --gpu={gpu_id} --label-ratio={label_ratio}

• Running all configurations:

  python src/main_SLeDGe_Light_full.py --dataset={dataset_name} --gpu={gpu_id} --label-ratio={label_ratio}

To apply temporal weighting, add -time to the end of any command:

python src/main_SLeDGe.py --dataset={dataset_name} --gpu={gpu_id} --label-ratio={label_ratio} -time

• The following table summarizes the statistics and sources for the benchmarks used in this framework.

• |f|, |c|, |D| are the number of input features, classes, and data size, respectively.