dbscan

Author: luoshao23

Clustering/DBSCAN.py

@@ -0,0 +1,67 @@
+import numpy as np
+from scipy.spatial.distance import cdist
+import matplotlib.pyplot as plt
+from sklearn.cluster import DBSCAN
+
+
+class cDBSCAN(object):
+    def __init__(self, min_pts=5, epsilon=0.5, metric='euclidean'):
+        self.min_pts = min_pts
+        self.epsilon = epsilon
+        self.metric = metric
+
+    def fit(self, X, y=None):
+        self.fit_predict(X, y)
+        return self
+
+    def predict(self, X):
+        pass
+
+    def fit_predict(self, X, y=None):
+        n_samples, _ = X.shape
+        nearin = cdist(X, X, metric=self.metric) <= self.epsilon
+        near_num = np.sum(nearin, axis=1)
+        core_ind = set(np.arange(n_samples)[near_num >= self.min_pts])
+        print(core_ind)
+
+        n_clusters = 0
+        this_set = set(range(n_samples))
+        clusters = []
+
+        while core_ind:
+            old_set = this_set.copy()
+            ele = core_ind.pop()
+            queue = [ele]
+            this_set.remove(ele)
+            while queue:
+                q = queue.pop(0)
+                if near_num[q] >= self.min_pts:
+                    dlt = this_set.intersection(np.arange(n_samples)[nearin[q]])
+                    queue.extend(dlt)
+                    this_set.difference_update(dlt)
+            n_clusters += 1
+            C = old_set.difference(this_set)
+            clusters.append(C)
+            core_ind.difference_update(C)
+        labels = -1 * np.ones(n_samples, dtype=int)
+
+        for l, g in enumerate(clusters):
+            labels[list(g)] = l
+        self.labels = labels
+        return labels
+
+if __name__ == "__main__":
+    np.random.seed(23)
+    X = np.random.random((40, 2))
+    cdb = cDBSCAN(epsilon=0.2)
+    res = cdb.fit_predict(X)
+    print(res)
+    db = DBSCAN(eps=0.2)
+    l = db.fit_predict(X)
+    print(l)
+
+    fig, (a1, a2) = plt.subplots(1, 2)
+
+    a1.scatter(X[:, 0], X[:, 1], c=res)
+    a2.scatter(X[:, 0], X[:, 1], c=l)
+    plt.show()