mlreco.utils.vertex module

mlreco.utils.vertex.find_closest_points_of_approach(point1, direction1, point2, direction2)[source]

Given two lines in 3D space, find the two points that are closest to each other on these lines.

See also https://math.stackexchange.com/a/1993990/391047.

Parameters

  • point1 (np.ndarray) – Point belonging to first line. Shape (3,)

  • direction1 (np.ndarray) – Direction defining the first line with point1. Shape (3,)

  • point2 (np.ndarray) – Point belonging to second line. Shape (3,)

  • direction2 (np.ndarray) – Direction defining the second line with point2. Shape (3,)

Output

tuple of np.ndarray – Two points of approach, tuple with shape ((3,), (3,))

mlreco.utils.vertex.get_ppn_points_per_particles(input_data, res, primary_particles, primary_particles_seg, data_idx=0, coords_col=(1, 4), attaching_threshold=10, track_label=1, shower_label=0, unwrapped=False, apply_deghosting=True, return_distances=False, min_voxel_count=10, min_track_count=2)[source]

Get predicted PPN points

Parameters

  • input_data (dict) –

  • res (dict) –

  • primary_particles (list) –

  • primary_particles_seg (list) –

  • data_idx (int, default 0) –

  • coords_col (tuple of int, default (1, 4)) –

  • attaching_threshold (float, default 10) – Distance (in voxels) to associate PPN candidates with particles.

  • track_label (int, default 1) – Semantic label for track-like particles.

  • shower_label (int, default 0) – Semantic label for shower-like particles.

  • unwrapped (bool, default False) –

  • apply_deghosting (bool, default True) – Whether PPN post-processing should consider the output already deghosted or not.

  • bool (return_distancesL) – For tuning studies.

  • False (default) – For tuning studies.

  • min_voxel_count (int, default 10) – Any particle with predicted deghosted voxel count below this threshold will be ignored.

  • min_track_count (int, default 2) – Will ignore shower particles as soon as we have >= min_track_count track-like particles in the interaction. This can be set to None to disable this behavior completely.

Output

  • - list of N arrays of shape (M_i,f) of M_i PPN candidate points, f corresponds to the number

  • of feature in the output of uresnet_ppn_type_point_selector.

  • - array of N lists of voxel indices, corresponding to the particles

  • whose predicted semantic is track or shower.

  • N is the number of particles which are either track or shower (predicted).

mlreco.utils.vertex.predict_vertex(inter_idx, data_idx, input_data, res, coords_col=(1, 4), primary_label=1, shower_label=0, track_label=1, attaching_threshold=10, inter_threshold=20, unwrapped=False, apply_deghosting=True, return_distances=False, other_primaries_threshold=10, other_primaries_gamma_threshold=100, pca_radius=28, min_track_count=2, min_voxel_count=10)[source]

Heuristic to find the vertex by looking at - predicted primary particles within predicted interaction - predicted PPN points for these primary particles

For now, very simple: taking the barycenter of potential candidates.

Parameters

  • inter_idx (int) – Predicted interaction index.

  • data_idx (int) – Batch entry index.

  • input_data (dict) – Input dictionary.

  • res (dict) – Output dictionary.

  • coords_col (tuple, default (1, 4)) –

  • primary_label (int, default 1) – In GNN predictions, integer tagging predicted primary particles.

  • shower_label (int, default 0) – Semantic label for shower-like particles.

  • track_label (int, default 1) – Semantic label for track-like particles.

  • attaching_threshold (float, default 10) – See get_ppn_points_per_particles.

  • inter_threshold (float, default 20) – PPN candidates need to minimize difference between distance to closest primary and distance of current primary particle voxels to closest primary particle.

  • unwrapped (bool, default False) – Whether input_data and res are already unwrapped or not.

  • apply_deghosting (bool, default True) – Whether to apply deghosting.

  • return_distances (bool, default False) – For tuning studies.

  • other_primaries_threshold (float, default 10) – Primaries too far from the other primaries will be ignored.

  • other_primaries_gamma_threshold (float, default 100) – Same as previous but for photon-like particles exclusively ($T_B$). Can be -1, then the same threshold as all other primaries will be used.

  • pca_radius (float, default 28) –

  • min_track_count (int, default 2) – See get_ppn_points_per_particles.

  • min_voxel_count (int, default 10) – See get_ppn_points_per_particles.

Output

np.ndarray – vtx_candidate with shape (3,)

mlreco.utils.vertex.get_vertex(kinematics, cluster_label, data_idx, inter_idx, vtx_col=9, primary_label=1)[source]

Getting true vertex for interaction identified by inter_idx

Look at kinematics label, selecting only primary particles within this interaction, and get vertex which occurs the most.

Parameters

  • kinematics (list of np.ndarray) – Kinematics labels.

  • cluster_label (list of np.ndarray) – Cluster labels.

  • data_idx (int) – Which entry we are looking at (labels).

  • inter_idx (int) – The true interaction id for which we want the vertex.

  • vtx_col (int, default 9) – First column of vertex coordinates in the kinematics labels. Coordinates columns go from vtx_col to vtx_col+2.

  • primary_label (int, default 1) – What integer tags primary particles in kinematics labels (“primary particles” ~ particles coming out of the vertex).

Output

np.ndarray – True vertex coordinates. Shape (3,)

mlreco.utils.utils module mlreco.visualization package