machine_learning

This file contains tools used for conducting machine learning tasks in Liquid Scintillator Detector. The purpose of 4 folders are listed below:

Preprocessing: Converting Monte Carlo Simulation .root file to a 2D Grid (theta, phi), and store them as a sparse matrix.

Network: Storing the Convolutional Neural Network Used to conduct the classification task, also include the hyperparameter search algorithm.

Analyzer: data analysis tool for making the pressure map and roc curves.

data: Data processing script to be use on BU SCC cluster.

Adding Spherical CNN feature modified from https://github.com/jonas-koehler/s2cnn

Please cite the following paper if you used this code:

DOI: 10.1016/j.nima.2019.162604

A full study contains:

Step 1 : Given Monte Carlo file, preprocess it using preprocessing/processing_sparse_time.py to convert it to a pmt hitmap.

Output : eventfile.pickles for each .root file
Step 2 : Collecting all eventfile.pickle into a list, each entry of the list should contains the address to the pickle files

Example(content in the .dat file) : /projectnb/snoplus/sphere_data/kamland_38/eventfile_sph_out_Xe136_dVrndVtx_3p0mSphere_1k_129.0.1000.pickle /projectnb/snoplus/sphere_data/kamland_38/eventfile_sph_out_Xe136_dVrndVtx_3p0mSphere_1k_122.0.1000.pickle /projectnb/snoplus/sphere_data/kamland_38/eventfile_sph_out_Xe136_dVrndVtx_3p0mSphere_1k_112.0.1000.pickle /projectnb/snoplus/sphere_data/kamland_38/eventfile_sph_out_Xe136_dVrndVtx_3p0mSphere_1k_1.0.1000.pickle

Output : There should be two .dat file in total: Xe136.dat C10.dat
Step 3 : Feeding .dat file into network/network_spherical.py. A set of pressure parameters(photocoverage, QE) will also need to be given in argument(by default 8,10, highest pressure and current KamLAND parameter)

Output : roc.png(plot of the ROC curve in validation data) sigmoid.png(plot of sigmoid output in validation data) roc_curve.npy(contain various numpy array of the output classification: (fpr, tpr, thr, sigmoid_s, sigmoid_b))
Step 4 : Making various plots. Plotting code is shown in the analyzer folder, e.g. roc_pm plots the pressure map.