############################################################################### # # # Data set used in Sadowski, Radics, Ananya, Baldi # # # # 'Efficient antihydrogen detection in antimatter physics by deep learning' # # # # Author: Peter Sadowski 2017 # # # ############################################################################### Files: 1) antihydrogen_train.csv.gz 2) antihydrogen_valid.csv.gz 3) antihydrogen_test.csv.gz 4) antihydrogen_vertex_train.csv.gz 5) antihydrogen_vertex_valid.csv.gz 6) antihydrogen_vertex_test.csv.gz Summary: The data set is divided into train, valid, and test sets, but also detector-level data (files 1-3) and vertex-level data (files 4-6), where the events (rows) in file 1 correspond to the events in file 4, etc. Detector-level data (Files 1-3): Each event in the detector level data consists of the target, 0 (on-axis annihilation) or 1 (on-wall annihilation), followed by 1430 binary values. These are organized as follows: Column 0: Target Columns 1-447: Inner-Z, 447 features. Columns 448-693: Inner-PHI, 246 features. Columns 694-1140: Outer-Z, 447 features. Columns 1141-1430: Outer-PHI, 290 features. Vertex-level data (Files 4-6): Each event in the vertex-level data files is described by the target, the number of valid vertices found by the vertex finding algorithm (possibly zero), and the coordinates and point-of-closest-approach of the best vertex (that with the smallest radius R=sqrt(x^2 + y^2)). Vertices returned by the vertex finding algorithm are required to have a POCA < 1.0 cm. Note that when a valid vertex was not found, the features of the best vertex are set to 0. Column 0: Target Column 1: Number of valid vertices found by the vertex finding algorithm. Column 2-4: X,Y,Z coordinates (in cm) of the best vertex. Column 5: Point of closest approach of the best vertex. This is inversely correlated with how accurate the vertex is.