Create dbscan_batch.py

exp/exp3/dcnn/snn/dbscan_batch.py

@@ -0,0 +1,81 @@
+from .query import *
+from .snn_cc import dfs1_cc, dfs2_cc, ndfs1_cc, ndfs2_cc, dfs3_cc
+from ..sample import *
+import numpy as np
+
+class dbscan2:
+
+    def __init__(self, algorithm='dbscan', eps=0.1, minPts=5, 
+                    metric="euclidean", init='uniform',
+                     sample_size=1000, memory_efficient=0, seed=42):
+
+        self.algorithm = algorithm
+        self.eps = eps
+        self.minPts = minPts
+        self.metric = metric
+        self.init = init
+        self.memory_efficient = memory_efficient
+        self.sample_size = sample_size
+        np.random.seed(seed)
+
+
+    def fit_transform(self, X):
+        self.fit(X)
+        return self.labels
+
+
+    def fit(self, X):
+
+        snnm = build_snn_model(X)
+        size = X.shape[0]
+
+        if self.algorithm == 'dbscan':
+
+            self.neighborhoods = snnm.radius_batch_query(X, 
+                                       self.eps,
+                                       return_distance=False)
+            self.neighborhoods = list(self.neighborhoods.values())
+            n_neighbors = np.asarray([len(neighbors) for neighbors in self.neighborhoods])
+            corePoints = np.asarray(n_neighbors >= self.minPts, dtype=np.uint8)
+
+            self.labels = np.full(size, -1, dtype=np.intp)
+            ndfs1_cc(corePoints, self.neighborhoods, self.labels)
+
+        elif self.algorithm == 'dbscan*':
+            self.neighborhoods = snnm.radius_batch_query(X, 
+                                       self.eps,
+                                       return_distance=False)
+
+            self.neighborhoods = list(self.neighborhoods.values())
+            n_neighbors = np.asarray([len(neighbors) for neighbors in self.neighborhoods])
+            corePoints = np.asarray(n_neighbors >= self.minPts, dtype=bool)
+            self.labels = np.full(size, -1, dtype=np.intp)
+            ndfs2_cc(corePoints, self.neighborhoods, self.labels)
+
+        elif self.algorithm == 'dbscan++':
+            if self.sample_size > size:
+                self.sample_size = int(round(0.5*size))
+
+            if self.init == 'uniform':
+                subsampleID = uniform_sample(X, size=self.sample_size)
+            else:
+                subsampleID = greedy_k_center_sample(X, size=self.sample_size)
+
+            subsampleID = uniform_sample(X, size=self.sample_size)
+            if self.memory_efficient:
+                self.neighborhoods = dict(snnm.radius_batch_query(X[subsampleID], self.eps, return_distance=False, memory_eff=1))
+            else:
+                self.neighborhoods = dict(snnm.radius_batch_query(X[subsampleID], self.eps, return_distance=False, memory_eff=0))
+
+            n_neighbors = np.array([len(self.neighborhoods[neighbors]) for neighbors in self.neighborhoods])
+            corePoints = np.zeros(size, dtype=np.uint8)
+            corePoints[subsampleID[n_neighbors >= self.minPts]]=1
+            self.labels = np.full(size, -1, dtype=np.intp)
+            dfs3_cc(corePoints, self.neighborhoods, subsampleID, self.labels)
+
+
+
+
+
+
+