Candidate Feature Lab

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.

Basic Pipeline

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.

1. 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');

2. Combine with HCTSA_Empirical1000.mat:

TS_Combine('HCTSA_Empirical1000.mat','HCTSA_newFeatures.mat',true,true,'HCTSA_merged.mat');

3. 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]);

Results

Test results for specific features are in the wiki.