DBSCAN classification for ESP32

This library is designed to be used with the Arduino IDE.

Dependencies

• no dependency

DBSCAN instance creation

First declare an instance of your DBSCAN:

dbscan DB(.6, 4, EUCLIDIAN);

The arguments are:

• epsilon: size of a local neighbourhood
• minimum number of points in a neighbourhood
• type of distance chosen:
  • EUCLIDIAN, // Euclidian distance
  • MINKOVSKI, // Minkowski distance (is Euclidian if param = 2) not working...
  • MANHATTAN, // Manhattan distance
  • CHEBYCHEV, // Chebychev distance
  • CANBERRA // Canberra distance

Dataset creation

A dataset is a vector of vectors of floats

std::vector<std::vector<float>> dataset

See the example on how to create a dataset. Basically, you need to declare the dataset as above and push all your vectors of float data into the dataset. E.g. if v is a vector of data of dimension 3:

std::vector<float> v;
float myV[] = {3, 3, 3};
v.assign (myV, myV + 3);
dataset.push_back(v);

Or more simply, if you have the data values

dataset.push_back([3,3,3]);

when the dataset is ready, just call the init method:

DB.init(dataset);

init returns the vector of clusters. The first vector is always the noise (even if empty), so the number of clusters created is equal to clusters.size() - 1

    std::vector<std::vector<uint16_t>> clusters;
    clusters = DB.init(Dataset);

The method displayStats provides usuful information about the clusters:

Created 3 clusters.
Cluster 0 : 70 points
    Centroid: 0.003065 0.004174 2.999999
    Tightness = 0.105
Cluster 1 : 56 points
    Centroid: 0.254542 0.028865 1.999999
    Tightness = 0.289
Cluster 2 : 60 points
    Centroid: -0.715484 0.040681 1.000000
    Tightness = 0.322

Separation = 1.517
Davies-Bouldin index = 0.383
14 noise points

Data prediction

To predict in which cluster a data is best located, use the predictmethod. Create the vector, send it to predict. The method returns the number of the best cluster or 65535 if your data is noise.

std::vector<float> v;
float myV[] = {3, 3, 3};
v.assign (myV, myV + 3);
uint16_t n = DB.predict(v);
Serial.printf("[3,3,3] in cluster %d\n", n);

Examples

Two examples are proposed, a simpleone to show the basics, and a more complex one with graphics that you can use on a TTGO T-display device. This example groups 2D points into 3 clusters (one in a circle in the center of the screen, and 2 around 2 circle arcs). You can use the buttons to change the values of epsilon and the tolerance of the points around the arcs, and GPIO15 to change the number of points in the dataset.