Splunk Machine Learning App / Toolkit - Using DBSCAN Clustering Algorithm

[hbrandt84]

Hi,

I want to use the Clustering Algorithm "DBSCAN" from the Machine Learning Toolkit.
(https://docs.splunk.com/Documentation/MLApp/2.3.0/User/Algorithms) --> listed under "clustering algorithms"

Now, upon implementation, I noticed, that this algorithm only needs one parameter: EPS
(maximum distance between two samples for them to be considered in the same cluster)

Now if you look up any definition of the DBSCAN Algorithm, for example...
(https://en.wikipedia.org/wiki/DBSCAN)
...you will notice that a DBSCAN algorithm will need 2 Parameters to be functional:

• EPS (Epsilon): maximum distance between two samples --> provided
• minPTS: minimum occurences of samples within a cluster --> missing

Does anybody know, why the second Parameter ist missing?
I Don't get how this algorithm can be functional....

[nryabykh]

You need to modify $SPLUNK_HOME/etc/apps/Splunk_ML_Toolkit/bin/algos/DBSCAN.py file. In __init__ function replace string

out_params = convert_params(options.get('params', {}), floats=['eps'])

with this one:

out_params = convert_params(options.get('params', {}), floats=['eps', 'min_samples'])

After this you can write something like fit DBSCAN eps=0.1 min_samples=2 * in your SPL queries.

[niketn]

@hbrandt84, I concur, scikit learn also mentions two parameters i.e. min_samples and eps (http://scikit-learn.org/stable/modules/clustering.html#dbscan)

However, algorithm description and class detail mention that these parameters are optional:
http://scikit-learn.org/stable/modules/generated/sklearn.cluster.DBSCAN.html

Based on the following code for DBSCAN algorithm, I would expect that initialization default value is min_samples=5 (https://github.com/scikit-learn/scikit-learn/blob/ab93d65/sklearn/cluster/dbscan_.py#L156):

def dbscan(X, eps=0.5, min_samples=5, metric='minkowski',
           algorithm='auto', leaf_size=30, p=2, sample_weight=None, n_jobs=1):

And:

def __init__(self, eps=0.5, min_samples=5, metric='euclidean',
             algorithm='auto', leaf_size=30, p=None, n_jobs=1):
    self.eps = eps
    self.min_samples = min_samples
    self.metric = metric
    self.algorithm = algorithm
    self.leaf_size = leaf_size
    self.p = p
    self.n_jobs = n_jobs

However, this needs to be confirmed and possibly enhanced in Machine Learning Toolkit to create a min_samples input parameter for DBSCAN.