Skip to content

Latest commit

 

History

History
31 lines (25 loc) · 1.3 KB

fcsets.md

File metadata and controls

31 lines (25 loc) · 1.3 KB
Raw

General Usage

When using FCSETS as stability measure for fuzzy over-time clusterings, first, FCSETS has to be initialized:

import pandas
import numpy
from ots_eval.stability_evaluation.fcsets import FCSETS

# test_data for 5 time series with 4 timestamps and 2 clusters per year
data = {
    1: numpy.array([[0.999, 0.9905, 0.989, 0.002, 0.028],
                    [0.001, 0.0005, 0.011, 0.998, 0.972]]),
    2: numpy.array([[0.999, 0.9905, 0.989, 0.002, 0.028],
                    [0.001, 0.0005, 0.011, 0.998, 0.972]]),
    3: numpy.array([[0.999, 0.9905, 0.989, 0.002, 0.028],
                    [0.001, 0.0005, 0.011, 0.998, 0.972]]),
    4: numpy.array([[0.999, 0.9905, 0.989, 0.002, 0.028],
                    [0.001, 0.0005, 0.011, 0.998, 0.972]])
}
rater = FCSETS(data)

FCSETS requires a dictionary with the form {<timestamp>: <membership_martix>} with membership_matrix being a 2-dim numpy.array of shape (num_clusters, num_time_series)

Now, the clustered data set can be evaluated with FCSETS.

clustering_score = rater.rate_clustering(start_time=None, end_time=None)

where start_time and end_time indicate the time intervall which should be considered. If start_time and end_time are None, the first and last timestamp are considered as boundary, respectively.