Created
January 16, 2018 11:16
-
-
Save barthr/f919080f2c3f2c493bec43da0a1010c0 to your computer and use it in GitHub Desktop.
Dbscan algorithm in python
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
import string | |
import numpy as np | |
UNCLASSIFIED = 0 | |
OUTLIER = -1 | |
class DBSCAN(object): | |
def __init__(self, epsilon=1.0, min_samples=10, dist=np.linalg.norm): | |
self.epsilon = epsilon | |
self.min_samples = min_samples | |
self.dist = dist | |
def fit_predict(self, X): | |
# we instantiate all the elements as unclassified | |
if len(X) == 0: | |
raise Exception("X is empty") | |
labels = len(X) * [UNCLASSIFIED] | |
cluster = 0 | |
for index in range(0, len(X)): | |
point = X[index] | |
if not labels[index] == UNCLASSIFIED: | |
continue | |
neighbours = self._region_query(point, X) | |
if len(neighbours) < self.min_samples: | |
labels[index] = OUTLIER | |
else: | |
cluster += 1 | |
self._expand_cluster(index, cluster, X, neighbours, labels) | |
return labels | |
def _region_query(self, point, X): | |
result = [] | |
for other in range(0, len(X)): | |
if self.dist(point - X[other]) < self.epsilon: | |
result.append(other) | |
return result | |
def _expand_cluster(self, index, cluster, dataset, neighbours, labels): | |
labels[index] = cluster | |
i = 0 | |
while i < len(neighbours): | |
current = neighbours[i] | |
if labels[current] == OUTLIER: | |
labels[current] = cluster | |
elif labels[current] == UNCLASSIFIED: | |
labels[current] = cluster | |
neighbour_of_neighbours = self._region_query(current, dataset) | |
if len(neighbour_of_neighbours) >= self.min_samples: | |
neighbours = neighbours + neighbour_of_neighbours | |
i += 1 | |
print(DBSCAN(epsilon=4.5, min_samples=1).fit_predict( | |
np.matrix('1 2;3 4;519 20; 47 50; 11 3;4 6;7 8;10.0 11;1.4 5.0'))) |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment