Asips for Pulsar Astronomy

'Asips' is a Research cunducted for automating pulsar candidate selection. This is the API of Asips which can be used by anyone. This implementation uses the HTRU2 dataset.

Quick links

• Website (Demo web app)
• Documentation

🔥 Features

Gaussian Hellinger Extremely Fast Decision Tree (GH-EFDT)

This is the main output of the research. GH-EFDT is a stream learning algorithm. This is an improviced version of Extremely Fast Decision Tree [1] for imbalanced streams. Hellinger Distance amoung Gaussian destributions were used as a split criterion [2] when handling the imbalanced problem. This algorithm is suitable for candidate selection since it is,

• 1. Accurate
• 2. Not biased toward the majority class
• 3. Learn incrementally
• 4. Not Harmed by concept drift
• 5. Fast

Other classification Algorithms

This API provide verification feature from other libraries. For now this provides,

• 1. OnlineSMOTEBaggingClassifier
• 2. OnlineUnderOverBaggingClassifier

🛠 Installation

Clone the repository

git clone https://github.com/Venoli/Asips-for-Pulsar-Astronomy.git

Run the Jupyter notebook inside the src folder

cd src/asips
!python flask_api.py

⚡️ Quick Start

Since this is developed using Flask, the above code will start the server on http://localhost:5000/. (This will refer as BASE_URL in the below sections)

Pretrain

Below request will pretrain the model.
count: pretrain count

BASE_URL/pretrain/<count>

Predict

Below request will make a prediction using the model.
count: number of samples to predict

BASE_URL/predict/<count>

Learn from all

By below request model will learn from all of the early predictions

BASE_URL/learn-from-all

Learn by id

By below request model will learn from sample with given id.
id: id of the sample

BASE_URL/learn/<id>

Test with Another Classifier

By below request previouse predictions can be verified using another model.
model: name of the model. (smoteBagging, underOverBagging)

BASE_URL/test-with-other-classifier/<model>

📖 Credits

• Extremely Fast Decission Tree (EFDT) [1] - GH-EFDT is a improved version of EFDT
• Hellinger Distance among Gaussian Distributions [2] - The improvemrnt done by using hellinger distance
• Scikit-Multiflow [3] - research, implimentation and testing was done on top of the scikit-multiflow library. scikit-multiflow implementation of EFDT was modified.
• Gaussian Hellinger Very Fast Decision Tree [4] - Main encouragement behind the GH-EFDT
• HTRU2 dataset [5] - The dataset that used in development

[1] C. Manapragada, G. I. Webb, and M. Salehi, “Extremely Fast Decision Tree,” 2018. DOI: 10.1145/nnnnnnn. arXiv: 1802.08780v1.

[2] R. J. Lyon, J. M. Brooke, J. D. Knowles, and B. W. Stappers, “Hellinger distance trees for imbalanced streams,” in Proceedings - International Conference on Pattern Recognition, Institute of Electrical and Electron- ics Engineers Inc., Dec. 2014, pp. 1969–1974, ISBN: 9781479952083. DOI: 10.1109/ICPR.2014.344. arXiv: 1405.2278.

[3] Montiel, J., Read, J., Bifet, A., & Abdessalem, T. (2018). Scikit-multiflow: A multi-output streaming framework. The Journal of Machine Learning Research, 19(72):1−5.

[4] R. J. Lyon, B. W. Stappers, S. Cooper, J. M. Brooke, and J. D. Knowles, “Fifty years of pulsar candidate selection: From simple filters to a new principled real- time classification approach,” Monthly Notices of the Royal Astronomical Society, vol. 459, no. 1, pp. 1104– 1123, Jun. 2016, ISSN: 13652966. DOI: 10.1093/mnras/ stw656. arXiv: 1603.05166.

[5] R. J. Lyon, B. W. Stappers, S. Cooper, J. M. Brooke, J. D. Knowles, Fifty Years of Pulsar Candidate Selection: From simple filters to a new principled real-time classification approach, Monthly Notices of the Royal Astronomical Society 459 (1), 1104-1123, DOI: 10.1093/mnras/stw656