search

IMPALA-Consortium / ctasval

https://impala-consortium.github.io/ctasval/
Other
0 stars 1 forks source link

anomaly_lof #3

Closed erblast closed 5 months ago

erblast commented 6 months ago

Follow design of: https://impala-consortium.github.io/ctasval/reference/anomaly.html

fcollinbi commented 5 months ago

Introduction of lof specific noise could be done by applying principal components to timeseries. Alter the first (most important) 1-5 components (I suggest starting out changing only 2 as those can be easily visualized in an 2d plot) and transform back. The noise could be something like adding y% of there average principal component value. Transformation is possible both ways as it is linear.

erblast commented 5 months ago

solved it now by sampling new data based on a random non-normal distribution for a percentage of patients determined by anomaly_degree