Loading

DISCORDS

gts outlier

The DISCORDS function detects discords in a Geo Time Series™. The discord detection algorithm is based on identifying outlier patterns in the Geo Time Series™. Outliers are patterns which differ significantly from their nearest neighbor.

The following constraints are enforced:

  • pattern length must divide window size
  • quantization scale must be a power of two from 2 to 65536
  • DISCORDS can only be applied to numeric Geo Time Series™ which are bucketized and filled (i.e. no bucket with missing values).
DISCORDS is available since version 1.0.0.

See also

Signatures

Examples

// // Create a GTS with 7200 datapoints and 0.1% of anomalies // NEWGTS 1 7200 <% DUP 2 * PI * 360 / SIN <% RAND 0.01 < %> <% 0.001 + %> IFT NaN NaN NaN 4 ROLL ADDVALUE %> FOR // // Apply DISCORDS // [ SWAP bucketizer.last 720 1 0 ] BUCKETIZE 32 // Window length 8 // Pattern length 16 // Quantization scale 5 // Maximum discord number false // Overlapping 0.0 // Distance ratio DISCORDS