Visualizing the data matrix

The clustered data matrix (if clustering has been performed, otherwise the non-clustered data matrix is used) can be visualized by running


This will produce a colored visualization of the data matrix such as that shown below.

When data is grouped according to a set of distinct keywords and stored as group metadata (using the TS_LabelGroups function), these can also be visualized using TS_plot_DataMatrix('colorGroups',1).

Visualizing the normalized (unclustered) data matrix

Running TS_plot_DataMatrix('norm') plots the data contained in the local file HCTSA_N.mat, yielding:

where black rectangles label missing values, and other values are shown from low (blue) to high (red) after normalization using the scaled outlier-robust sigmoidal transformation. Due to the size of the matrix, operations are not labeled.

Examples of time series segments are shown to the left of the plot, and when the middle plot is zoomed, the time-series annotations remain matched to the data matrix:

Visualizing the clustered data matrix

It can be useful to display the matrix with the order of time series and operations preserved, but the relationships between rows and columns can be difficult to visualize when ordered randomly.

By running TS_plot_DataMatrix('cl'), the clustering information contained in HCTSA_N.mat is used to reorder the time series and features by similarity (if this information exists, cf. TS_cluster), yielding:

By reordering rows and columns, this representation reveals correlated patterns of outputs across different types of operations, and similar sets of properties between different types of time series.

Example: Incorporating group information

In this example, we consider a set of 20 periodic and 20 noisy periodic signals. We assigned the time series to groups (using TS_LabelGroups('orig',{'periodic','noisy'},'ts')), normalized the data matrix (TS_normalize), and then clustered it (TS_cluster). So now we have a clustered data matrix containing thousands of summaries of each time series, as well as pre-assigned group information as to which time series are periodic and which are noisy. When the time series have been assigned to groups , this can be accessed by setting the second input to 1:

TS_plot_DataMatrix('cl','colorGroups',false); % don't color according to group labels
TS_plot_DataMatrix('cl','colorGroups',true); % color according to group labels

producing the following two plots:

When group information is not used (the left plot), the data is visualized in the default blue/yellow/red color scheme, but when the assigned groups are colored (right plot), we see that the clustered dataset separates perfectly into the periodic (green) and noisy (blue) time series, and we can visualize the features that contribute to the separation.