hctsa

Mostly small tweaks. Some keywords updated, and catch22 short names added.

• New partial autocorrelation features up to 20 lags, and the RAD feature of @brendanjohnharris
• Some clearer and better-commented syntax
• catch22 is now a git submodule in Toolboxes, rather than being embedded.
• Feature-set lists more clearly distinguished, e.g., as INP_mops_hctsa.txt and INP_ops_hctsa.txt (versus equivalent files for _catch22)

• TS_SimSearch supports interactive plots to swap between raw and ranked values in scatter plots
• Better error messaging for computation
• install script checks for required toolboxes
• 'svmBeta' option to score individual features on SVM beta weights: added to TS_TopFeatures
• Better checks on input data type (e.g., singles or integers) in feature computation
• Improved reporting of in-fold and out-of-fold accuracies in k-fold CV classification
• Logistic regression added to classifiers, as well as svm-linear (high-dim) default (original available as svm-linear-lowdim)
• TS_ClassifyLowDim expanded functionality to check for in-sample over-fitting
• Null testing improved with option for simple shuffle-based nulls (under assumption that random in -> random out; rather than full model-based nulls)
• Improved visualization settings in TS_TopFeatures (including Spearman correlations now default)
• signed p-value test statistic from rank-sum test added in TS_TopFeatures
• Improved syntax for TS_Init for selecting a feature set by name.
• Progress bar option for new minimal output versions of TS_Compute as 'fast' and 'minimal' (in cases like with catch22 where full output makes the commandline unreadable.
• Some feature keywords improved.

• Interactive plotting (annotations added in response to mouse clicks) in TS_LowDimInspect and TS_FeatureSummary
• Histogram binning now done in a simpler and more transparent way, to match C implementation in catch22 (in DN_HistogramMode).
• New histogram asymmetry features added via DN_HistogramAsymmetry.

• Clearer naming of the CO_HistogramAMI features
• New nsadf feature
• New features of first maximum of various self-correlation function (rather than just first min) added.
• Various additional fixes

• Clearer naming of CO_HistogramAMI features
• Ability to specify reduced feature subsets in cfnParams (that will be applied in TS_Classify, TS_ClassifyLowDim, TS_TopFeatures, TS_PlotLowDim)
• Clearer keyword labeling of 'locDep' -> 'locationDependent', etc.

Added continuous analogues of first crossings

CO_FirstZero and CO_f1ecac combined into CO_FirstCrossing, and added discrete (first time lag the ACF exceeds the threshold) and continuous (linear interpolation for the crossing estimate, between previous and subsequent time points).
Also an issue with BF_PreProcess consistency and with 'absclose' being incorrectly set to 'saturate' causing an issue for DN_RemovePoints

Consistent capitalization across functions, and clearer convention around z-scoring as x_z rather than y.

• DN_RemovePoints has some new options, with either 'remove' or 'saturate'.
• Improved behavior and statistics from CO_AutoCorrShape

Tweaked some features, fixed some errors. Semantic versioning practices are not quite valid for hctsa's feature library (where every slight change to any feature makes hctsa results inconsistent with every previous version). But it's about time we had a v1.0!

Cleaning of poor-performing features, FourierPowerSpectrum keyword used, TS_plot_timeseries -> TS_PlotTimeSeries for consistent naming convention, improvement in Gaussian mutual information computation.

• Metadata for TimeSeries, Operations, and MasterOperations are now stored in tables (rather than clunkier cell arrays).
• TS_plot_pca renamed TS_PlotLowDim and now supports tSNE projections.
• Added very simple function for reducing the number of features as TS_ReduceFeatureSet.

hctsa version used in: Fulcher, B. D., & Jones, N. S. (2017). hctsa: A Computational Framework for Automated Time-Series Phenotyping Using Massive Feature Extraction. Cell Systems, 5(5), 527–531.e3.

Major working version.