Carolina Environmental Program UNC Chapel Hill

1
The Analysis Engine A New Tool for Model
Evaluation, Sensitivity and Uncertainty Analysis,
and more

• Alison M. Eyth, Prashant P. Pai
• Carolina Environmental Program
• University of North Carolina at Chapel Hill
• October 19, 2004

2
Background

• Supports data analysis by creating plots and
  tables
• Analysis Configurations facilitate repeated
  analyses
• Developed as part of the Multimedia Integrated
  Modeling System (but can be used standalone)
• Java application that runs on Windows, Linux,
• Open source available from http//sourceforge.ne
  t/projects/mimsfw
• Three main components
• Table application
• Plotting engine
• Statistics package

3
Table Application

• Provides the top level user interface
• File menu accesses import and export functions
• Currently supported file formats include
• Comma separated (.csv), Custom and tab delimited,
  Fixed column width, SMOKE Report, and ARFF
• Data files are imported as rows and columns
• Each file is shown in its own tab with file name,
  header, data table, and footer
• Toolbar and popup menus provide access to
  functions (e.g. sort, filter, format, plot,
  statistics)

4
Table Application GUI
5
Toolbar and Pop-up Menu Functions

• Multi-column sort
• Show rows with Top N values
• Show tows with Bottom N values
• Filter rows based on criteria (e.g. NOx gt 500)
• Show / hide columns
• Format columns (e.g. number style, color, width)
• Create plots
• Compute statistics
• Edit analysis configuration
• Reset

6
Filter Rows Dialog

• Use Filter Rows to limit the rows shown in the
  table
• Any number of criteria can be added
• Each criterion has a column, operation, and value
• Available operations are lt, lt, gt, gt, not ,
  starts with, contains, ends with, does not start
  with, does not contain, ...
• Select between showing rows matching ALL criteria
  or ANY

7
Plotting Options Dialog

• Choose Plot type from Bar, Box, CDF, Discrete
  Category, Histogram, Rank Order, XY, Line, Time
  Series, and Tornado
• Select Data Columns to plot
• Specify Units and one to three columns to use for
  labels
• Selected data is passed to the plotting engine

8
Plot Properties are Specified using the Analysis
Engine GUI
9
Example Discrete Category Plot
Note Plots are created using a custom Java
interface to R
10
Statistics Dialog

• Provides interface to the statistics package
• Specify statistics to compute and data columns to
  analyze
• Additional details are specified on other tabs
• Statistics outputs appear as new tabs in the
  table application
• Statistics are computed using Colt and Weka

11
Example of Histogram Statistics
12
Analysis Configuration Dialog

• The Analysis Configuration stores all the table
  settings and plots that you have created during
  your session
• The selected plots can be viewed, edited or
  deleted
• Plots can be given new names by double clicking
  the name
• Some (or all) of the settings can be saved to a
  configuration file
• Configuration files can be loaded in future
  sessions or for other data files in the current
  session

13
Automation

• An optional command line interface may be used
  specify
• Data files to load
• Analysis configuration file to use
• Type of plots to create (e.g. JPG, PDF, PNG)
• Output directory for plots and tables
• This allows plots and tables to be created in an
  automated fashion
• Standard analysis products may be created for
  newly available data sets

14
Examples of Potential Applications

• Model Evaluation
• Sort to find stations at which the error was the
  largest
• Plot modeled and observed values on box plots,
  etc.
• Create scatter plots of one species vs. another
• Sensitivity and Uncertainty Analysis
• Perform linear regression and show in plots and
  tables
• Compute correlation coefficients
• Emissions Modeling Quality Assurance
• Find states with top 10 emission values
• Stacked bar charts to show total emissions
• Compute histograms
• General Data Analysis
• Analyze data by sorting, filtering, and computing
  statistics

15
Future Directions

• Initial version will be released on SourceForge
  by 10/31/04 (which is the end date for the
  current funding for this work)
• Many potential enhancements are listed on
  SourceForge, e.g.
• Create new rows and columns using functions (e.g
  difference, sum)
• Create plots and tables with data from multiple
  tabs
• Will likely be used as part of the new emissions
  quality assurance tool (http//sourceforge.net/pro
  jects/emisview)
• Mr. Tommy Cathey will continue to develop the
  custom Java interface to R at the EPA Scentific
  Visualization Laboratory in FY05

16
References

• MIMS Sourceforge page (for downloads)
  http//sourceforge.net/projects/mimsfw
• R (for plots) http//www.r-project.org
• Colt (for basic statistics) http//www-itg.lbl.g
  ov/hoschek/colt
• Weka (for regression and correlation analysis)
  http//www.cs.waikato.ac.nz/ml/weka/
• Carolina Environmental Program (for more
  information) http//www.cep.unc.edu
• Primary Author eyth_at_unc.edu