fixed issue with num_processes at centroid_calculator

pymepix/main.py

@@ -367,7 +367,7 @@ def main():
         "--number_of_processes",
         dest="number_of_processes",
         type=int,
-        default=1,
+        default=4,
         help="The number of processes used for the centroiding (default: 1 => parallel processing disabled')",
     )
 

pymepix/processing/logic/centroid_calculator.py

@@ -22,6 +22,8 @@
 from threading import current_thread
 from time import time
 
+import os
+
 import numpy as np
 import scipy.ndimage as nd
 from sklearn.cluster import DBSCAN
@@ -159,7 +161,7 @@ class CentroidCalculator(ProcessingStep):
     def __init__(
         self,
         cent_timewalk_lut=None,
-        number_of_processes=4,
+        number_of_processes=1,
         clustering_args={},
         dbscan_clustering = True,
         *args,
@@ -386,12 +388,10 @@ def centroid_chunks_to_centroids(self, chunks):
     def perform_centroiding_dbscan(self, chunks):
 #        with Pool(self.number_of_processes) as p:
 #            return p.map(self.calculate_centroids_dbscan, chunks)
-
-        if self.number_of_processes>1:
-#            return Parallel(n_jobs=self.number_of_processes)(delayed(calculate_centroids_dbscan)(c,self.tot_threshold, self._tof_scale, self.epsilon, self.min_samples, self._cent_timewalk_lut) for c in chunks)
-            return Parallel(n_jobs=None)(delayed(calculate_centroids_dbscan)(c,self.tot_threshold, self._tof_scale, self.epsilon, self.min_samples, self._cent_timewalk_lut) for c in chunks)
-
-        return map(self.calculate_centroids_dbscan, chunks)
+
+        return Parallel(n_jobs=self.number_of_processes)(delayed(calculate_centroids_dbscan)(c,self.tot_threshold, self._tof_scale, self.epsilon, self.min_samples, self._cent_timewalk_lut) for c in chunks)
+
+#        return map(self.calculate_centroids_dbscan, chunks)
 
     def perform_centroiding_cluster_stream(self, chunks):
         self.cstream = ClusterStream(self.cs_sensor_size, self.cs_max_dist_tof,\
@@ -448,89 +448,8 @@ def calculate_centroids_cluster_stream(self, chunk):
 
         return None
 
-    def perform_clustering_dbscan(self, shot, x, y, tof):
-        """ The clustering with DBSCAN, which is performed in this function is dependent on the
-            order of the data in rare cases. Therefore, reordering in any means can
-            lead to slightly changed results, which should not be an issue.
-
-            Martin Ester, Hans-Peter Kriegel, Jiirg Sander, Xiaowei Xu: A Density Based Algorithm for
-            Discovering Clusters [p. 229-230] (https://www.aaai.org/Papers/KDD/1996/KDD96-037.pdf)
-            A more specific explaination can be found here:
-            https://stats.stackexchange.com/questions/306829/why-is-dbscan-deterministic"""
-        if x.size >= 0:
-            X = np.column_stack(
-                (shot * self.epsilon * 1_000, x, y, tof * self._tof_scale)
-            )
-
-            dist = DBSCAN(
-                eps=self.epsilon,
-                min_samples=self.min_samples,
-                metric="euclidean",
-                n_jobs=1,
-            ).fit(X)
 
-            return dist.labels_ + 1
-
-        return None
-
-
-    def calculate_centroids_properties(self, shot, x, y, tof, tot, labels):
-        """
-        Calculates the properties of the centroids from labeled data points.
-
-        ATTENTION! The order of the points can have an impact on the result due to errors in
-        the floating point arithmetics.
-
-        Very simple example:
-        arr = np.random.random(100)
-        arr.sum() - np.sort(arr).sum()
-        This example shows that there is a very small difference between the two sums. The inaccuracy of
-        floating point arithmetics can depend on the order of the values. Strongly simplified (3.2 + 3.4) + 2.7
-        and 3.2 + (3.4 + 2.7) can be unequal for floating point numbers.
-
-        Therefore there is no guarantee for strictly equal results. Even after sorting. The error we observed
-        can be about 10^-22 nano seconds.
 
-        Currently this is issue exists only for the TOF-column as the other columns are integer-based values.
-        """
-        label_index, cluster_size = np.unique(labels, return_counts=True)
-        tot_max = np.array(
-            nd.maximum_position(tot, labels=labels, index=label_index)
-        ).flatten()
-
-        tot_sum = nd.sum(tot, labels=labels, index=label_index)
-        tot_mean = nd.mean(tot, labels=labels, index=label_index)
-        cluster_x = np.array(
-            nd.sum(x * tot, labels=labels, index=label_index) / tot_sum
-        ).flatten()
-        cluster_y = np.array(
-            nd.sum(y * tot, labels=labels, index=label_index) / tot_sum
-        ).flatten()
-        cluster_tof = np.array(
-            nd.sum(tof * tot, labels=labels, index=label_index) / tot_sum
-        ).flatten()
-        cluster_totMax = tot[tot_max]
-        cluster_totAvg = tot_mean
-        cluster_shot = shot[tot_max]
-
-        if self._cent_timewalk_lut is not None:
-            # cluster_tof -= self._timewalk_lut[(cluster_tot / 25).astype(np.int) - 1]
-            # cluster_tof *= 1e6
-            cluster_tof -= (
-                self._cent_timewalk_lut[np.int_(cluster_totMax // 25) - 1] * 1e3
-            )
-            # TODO: should totAvg not also be timewalk corrected?!
-            # cluster_tof *= 1e-6
-
-        return (
-            cluster_shot,
-            cluster_x,
-            cluster_y,
-            cluster_tof,
-            cluster_totAvg,
-            cluster_totMax,
-            cluster_size,
-        )
 
 
 class CentroidCalculatorPooled(CentroidCalculator):