Merge pull request #162 from jacanchaplais/patch

Parallel computation for clustering coefficients

graphicle/calculate.py

@@ -559,12 +559,12 @@ def _delta_R_symmetric(
     return result
 
 
-@nb.njit("float32[:](bool_[:, :])")
+@nb.njit("float32[:](bool_[:, :])", parallel=True)
 def _clust_coeffs(adj: base.BoolVector) -> base.FloatVector:
     num_nodes = adj.shape[0]
     coefs = np.empty(num_nodes, dtype=np.float32)
-    for node_idx, row in enumerate(adj):
-        neibs = np.flatnonzero(row)
+    for node_idx in nb.prange(num_nodes):
+        neibs = np.flatnonzero(adj[node_idx, :])
         num_neibs = neibs.shape[0]
         if num_neibs < 2:
             coefs[node_idx] = 0.0
@@ -575,7 +575,10 @@ def _clust_coeffs(adj: base.BoolVector) -> base.FloatVector:
 
 
 def cluster_coeff_distbn(
-    pmu: "MomentumArray", radius: float, pseudo: bool = True
+    pmu: "MomentumArray",
+    radius: float,
+    pseudo: bool = True,
+    threads: int = 1,
 ) -> base.FloatVector:
     """A measure of clustering for a particle point-cloud. Transforms
     point-cloud into a graph, where node neighbourhood is determined by
@@ -596,15 +599,19 @@ def cluster_coeff_distbn(
     pseudo : bool
         If True, will use pseudorapidity, rather than true rapidity.
         Default is True.
+    threads : int
+        Number of threads to use in the parallel portions of the
+        calculation.
 
     Returns
     -------
     ndarray[float32]
         Clustering coefficients of the particles in the point-cloud.
     """
-    adj = pmu.delta_R(pmu, pseudo) < radius
+    adj = pmu.delta_R(pmu, pseudo, threads) < radius
     np.fill_diagonal(adj, False)
-    return _clust_coeffs(adj)
+    with _thread_scope(threads):
+        return _clust_coeffs(adj)
 
 
 @ctx.contextmanager

graphicle/select.py

@@ -11,11 +11,11 @@
 import typing as ty
 
 import awkward as ak
-import fastjet
+import fastjet as fj
 import numpy as np
 import pandas as pd
 import scipy.optimize as opt
-from scipy.sparse.csgraph import breadth_first_tree
+from scipy.sparse.csgraph import breadth_first_tree as _breadth_first_tree
 
 import graphicle as gcl
 
@@ -113,21 +113,22 @@ def fastjet_clusters(
     pmu_ak = ak.from_numpy(pmu_renamed)
     extra_param = tuple()
     if p_val == -1:
-        algo = fastjet.antikt_algorithm
+        algo = fj.antikt_algorithm
     elif p_val == 0:
-        algo = fastjet.cambridge_algorithm
+        algo = fj.cambridge_algorithm
     elif p_val == 1:
-        algo = fastjet.kt_algorithm
+        algo = fj.kt_algorithm
     else:
-        algo = fastjet.genkt_algorithm
+        algo = fj.genkt_algorithm
         extra_param = (p_val,)
-    jetdef = fastjet.JetDefinition(algo, radius, *extra_param)
-    sequence = fastjet.ClusterSequence(pmu_ak, jetdef)
+    jetdef = fj.JetDefinition(algo, radius, *extra_param)
+    sequence = fj.ClusterSequence(pmu_ak, jetdef)
     jets = sequence.inclusive_jets()
     jet_pmus = gcl.MomentumArray(jets.to_numpy())
+    num_jets = len(jet_pmus)
     pt_descend_idxs = np.argsort(jet_pmus.pt)[::-1].tolist()
     jet_pmus = jet_pmus[pt_descend_idxs]
-    cuts = np.ones(len(pt_descend_idxs), dtype=np.bool_)
+    cuts = np.ones(num_jets, dtype=np.bool_)
     if pt_cut is not None:
         cuts = jet_pmus.pt > pt_cut
     if eta_cut is not None:
@@ -263,7 +264,7 @@ def vertex_descendants(adj: gcl.AdjacencyList, vertex: int) -> gcl.MaskArray:
         return gcl.MaskArray(adj.edges["dst"] == vertex)
     sparse = adj._sparse_signed
     vertex = sparse.row[vertex_mask][0]
-    bft = breadth_first_tree(adj._sparse_csr, vertex)
+    bft = _breadth_first_tree(adj._sparse_csr, vertex)
     mask = np.isin(sparse.row, bft.indices)
     mask[sparse.row == vertex] = True  # include edges directly from parent
     return gcl.MaskArray(mask)