11
$\begingroup$

I would like to use the knn distance plot to be able to figure out which eps value should I choose for the DBSCAN algorithm. Based on this page:

The idea is to calculate, the average of the distances of every point to its k nearest neighbors. The value of k will be specified by the user and corresponds to MinPts. Next, these k-distances are plotted in an ascending order. The aim is to determine the “knee”, which corresponds to the optimal eps parameter.

Using python with numpy/sklearn, I have the following points, with the following distance for 6-knn:

X = np.array([[-1, -1], [-2, -1], [-3, -2], [1, 1], [2, 1], [3, 2]])
nbrs = NearestNeighbors(n_neighbors=len(X)).fit(X)
distances, indices = nbrs.kneighbors(X)

# Indices

[[0 1 2 3 4 5]
 [1 0 2 3 4 5]
 [2 1 0 3 4 5]
 [3 4 5 0 1 2]
 [4 3 5 0 1 2]
 [5 4 3 0 1 2]]

# Distances
[[ 0.          1.          2.23606798  2.82842712  3.60555128  5.        ]
[ 0.          1.          1.41421356  3.60555128  4.47213595  5.83095189]
[ 0.          1.41421356  2.23606798  5.          5.83095189  7.21110255]
[ 0.          1.          2.23606798  2.82842712  3.60555128  5.        ]
[ 0.          1.          1.41421356  3.60555128  4.47213595  5.83095189]
[ 0.          1.41421356  2.23606798  5.          5.83095189  7.21110255]]

then I computed the average distance:

distances.mean()
2.9269575028354495

The problem is I don't understand how exactly could I represent the same plot as them with distances in y-axis and number of points according to the distances on the x-axis using python.

Thank for your help.

Improve this question
$\endgroup$
2

1 Answer 1

Reset to default
9
$\begingroup$

You

  1. take the last column of that matrix
  2. sort descending
  3. plot index, distance
  4. hope to see a knee (if the distance does not work well. there might be none)
Improve this answer
$\endgroup$
6
  • $\begingroup$ On the same plot, I do this for different k? or only one k for one plot as in the example? and what do you mean by "index" $\endgroup$
    – Marc Lamberti
    Commented Feb 9, 2016 at 20:53
  • $\begingroup$ Using the 6NN when you only have 6 points is of course nonsense. Do it for an appropriate k. Index as in "array index". because you need 2d to plot. $\endgroup$
    – Has QUIT--Anony-Mousse
    Commented Feb 9, 2016 at 20:57
  • $\begingroup$ And i only use the last column of the distance matrix. Because in the example they talk about averaging distances.. $\endgroup$
    – Marc Lamberti
    Commented Feb 9, 2016 at 22:26
  • $\begingroup$ That post is incorrect there and in at least another place (you don't need to set a seed) $\endgroup$
    – Has QUIT--Anony-Mousse
    Commented Feb 9, 2016 at 22:46
  • 1
    $\begingroup$ You only have one k. Why don't you use the DBSCAN paper. but mash-up various low-quality websites? $\endgroup$
    – Has QUIT--Anony-Mousse
    Commented Feb 9, 2016 at 22:53

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.