edgeslab/sampling-osal

Effectiveness of Sampling Strategies for One-shot Active Learning from Relational Data ~ MLG'20

One-shot Active Learning from Relational Data

Requirements

• This project requires python 3.6
• Python package requiremnts is listed in requirements.txt

Preparing environment

• We highly recomment using virtual environment for this part of the project. To create a virtual environment with Anaconda and activating it, run the following:

    conda create -n osal python=3.6
    conda activate osal

• Then install all the dependencies though pip:

    pip install -r requirements.txt

Preparing data

In order to download the datasets and process it for experiments, first run the following script once:

    sh prep.sh

Then follow the instruction inside the script for further data processing

Usage

To reproduce the results in MLG paper, run the following:

    sh mlg20.sh <num_trials> <budget>

• Use the following values for the parameters:
  • num_trials = 5
  • budget = 224

In order to run a single experiment, execute following:

    python experiment.py -config configs/<clf>_<ds>.json -nt <nt> -b <b> -algos <algos>

• Parameter description:

  • <clf> : classfier name (e.g. wvrn, cc, sgc, gsage)
  • <ds> : dataset name (e.g. citeseer, cora, hateful, pubmed)
  • <nt> : number of trials (e.g. 1, 5)
  • <b> : active learning budget (e.g. 32, 224)
  • <algos> : space separated names of sampling algo (e.g. rs, fp, ffs .. )

• A simple run:

    python experiment.py -config configs/wvrn_cora.json -nt 1 -b 32 -algos rs fp

---

Citation

@inproceedings{ahsan-mlg20,
  title={Effectiveness of Sampling Strategies for One-shot Active Learning from Relational Data},
  author={Ahsan, Ragib and Zheleva, Elena},
  booktitle={Proceedings of the 16th International Workshop on Mining and Learning with Graphs (MLG)},
  year={2020}
}