Time Series Parameters Dialog

The next screen of the wizard will depend on which data format you selected. If you selected Xyce or CSV, the next screen will be a file browser on the remote host. Navigate to the location of the ensemble data table (a dakota_tabular file within the directory hierarchy described above for Xyce inputs, or a CSV file containing full paths to each time series file for CSV inputs). Navigation is identical to that described in the earlier section on Remote Files. Click on the data table file in the remote file browser to select it, then click Continue. If you selected HDF5, the wizard skips this step since there is no need to select a table file.

../_images/Figure50.png

Figure 50: Timeseries Parameters for Xyce data sets.

For all three input types, the next step is setting the parameters to be used for binning and clustering the time series. Xyce, CSV, and HDF5 have slightly different interfaces for this step, which are shown in Figure 50, Figure 51, and Figure 52, respectively.

../_images/Figure51.png

Figure 51: Timeseries Parameters for CSV data sets.

../_images/Figure52.png

Figure 52: Timeseries Parameters for HDF5 data sets.

The CSV screen includes two additional fields that are not needed by the other formats, Table File Delimiter and Timeseries Column Name. Table File Delimiter allows you to use other delimiters besides commas in the data table, such as tabs or spaces. Tabs are difficult to specify because the web interface uses tabs to move between fields, but if you cut-and-paste a tab into the field, enclosing it with single quotes, Slycat™ will accept a tab-delimited table. To designate a space as a delimiter, enclose it with single quotes, since otherwise the field is interpreted as being empty. Commas do not require quotes.

Since CSV data tables can have multiple columns of time series data (e.g. if you sampled a set of variables over time at various locations within the simulation), the Timeseries Column Name identifies which time series data set to analyze. Type in the column name, taking care to exactly match the header as it appears in the table.

The remaining parameters are shared by all three input types. Timeseries Bin Count controls how finely the time series is sampled. The resulting binned sequences are used for calculating similarities and the reduced representations are drawn in the model visualization. Generally, bin counts between 500 and 1000 produce a reasonable tradeoff between speed and accuracy. Although increasing the number of bins increases both the analysis and rendering times, a greater bin count also helps preserve spikes or other localized features that could be lost when using a smaller number.

The Resampling Algorithm dropdown has two options. Both algorithms use a uniform set of bins, with the choice between using uniform piecewise linear approximation or uniform piecewise aggregate approximation as the resampling method. Uniform piecewise aggregate approximation is the default.

The Cluster Linkage Measure dropdown selects the metric used when evaluating distance between groups of elements. There are four choices:

  • single: Nearest Point Algorithm

  • complete: Farthest Point Algorithm

  • average: Unweighted Pair Group Method with Arithmetic Mean (UPGMA) Algorithm

  • weighted: Weighted Pair Group Method with Arithmetic Mean (WPGMA) Algorithm

Single evaluates the distance using the closest elements/minimum linkage; complete uses the farthest elements/maximum linkage; average uses the distance between the group averages; and weighted uses the values from the distance matrix. Average is the default. We are using SciPy to perform the clustering, so a more complete description of the linkage choices can be found at https://docs.scipy.org/doc/scipy/reference/generated/scipy.cluster.hierarchy.linkage.html.

The Cluster Metric currently only has a single choice, Euclidean, so it cannot be changed (hence the field is grayed out). This field is provided to inform you that we are using Euclidean distances in our algorithms.

Once you are satisfied with these parameter choices, click Continue to go to the next screen.