This is about testing newly proposed features for inclusion in hctsa. Features that are sufficiently unique, when judged across a diverse range of time series are included.
See the hctsa gitbooks page for more detailed instructions.
First, download HCTSA_Empirical1000.mat
and INP_1000ts.mat
from figshare, as an example data context for comparison.
- Prepare input files:
INP_master_ops_new.txt
,INP_ops_new.txt
and compute their features (using the same version of hctsa as you downloaded the precomputed versions from figshare):
TS_Init('INP_1000ts.mat','INP_master_ops_new.txt','INP_ops_new.txt',false,'HCTSA_newFeatures.mat');
TS_Compute(false,[],[],'missing','HCTSA_newFeatures.mat');
- Combine with
HCTSA_Empirical1000.mat
:
TS_Combine('HCTSA_Empirical1000.mat','HCTSA_newFeatures.mat',true,true,'HCTSA_merged.mat');
- Check behavior of new feature (e.g.,
myNewFeature
):
load('HCTSA_merged.mat','Operations');
theID = Operations.ID(strcmp(Operations.Name,'myNewFeature'));
TS_SimSearch(theID,'tsOrOps','ops','numNeighbors',40,'whatData','HCTSA_merged.mat','whatPlots',{'scatter','matrix'})
You can also run the specific file in this repository, e.g.,:
loadedData = load('HCTSA_merged.mat');
TS_FeatureFeatureScatter(loadedData,[871,7704]);
Test results for specific features are in the wiki.