ALSIM: Active Learning Simulator

ALSIM is an active learning strategy simulator. Active learning is a sub-domain of machine learning focused on the creation of machine learning models using the lowest amount of annotated data. This is especially interesting for applications where data acquisition and/or annotation is difficult, expensive and/or time-consuming. This simulator is able to test newly created active learning strategies and directly compare them to other algorithms.

AWUS is a novel, and PATENTED state-of-the-art active learning query strategy, outperforming all other strategies currently implemented, at very low computational cost. The journal paper can be found at AWUS: Adaptive Weighted Uncertainty Sampling.

Main Features / Comments

Major information:

• Multiple query strategies available out-of-the-box
• All Scikit-Learn machine learning models supported
• Fast custom ML models which are optimized for Active Learning available.
• Visualization build in.

How to get it

Git has to be installed to clone:

sudo apt install git

Clone the repository to current working directory

git clone https://github.com/gijsvanhoutum/alsim.git

We advise to install a new python virtual environment first with:

python3 -m venv venv

Activate environment

source venv/bin/activate

Install all necessary Python packages with:

pip install -r /alsim/requirements.txt

How to run it

To run execute the following from the current working directory:

python3 run_simulations.py

TODO

• Expand capabilities which is supported by the ALSIM AWUS paper version.