attpc-event-classification

Evaluating Machine Learning Methods for Event Classification in the Active-Target Time Projection Chamber

This work is a survey of methods to use for track classification in the AT-TPC. The work was done with the goal of classifying proton tracks from the 46Ar(p,p) experiment that ran in August of 2015.

This repository contains code produced for Jack Taylor's 2017-18 academic year independent research project. All results found in this work are presented in my physics honors thesis, and are also available on the arXiv: https://arxiv.org/abs/1810.10350.

Algorithms tested include those available in the scikit-learn package and neural networks written using Keras with a Tensorflow backend.

Dependencies / Packages used

• pytpc
• numpy
• matplotlib
• scipy
• pandas
• scikit-learn
• keras
• tensorflow

See requirements.txt for more exhaustive list with release information.

Models/Algorithms Explored

• Logistic Regression
• Single-Layer Densely-Connected Neural Network
• Two Layer Densely-Connected Neural Network
• Pre-Trained Convolutional Neural Network (VGG16 Architecture - Image Recognition Problem)
• Support Vector Machines (One Class Classification)