GoldSim

1

GoldSim Tutorial and Discussion Techniques for
Testing, Ensuring Quality and Validating GoldSim
Models
GoldSim User Conference San Francisco, CA October
25-26, 2007
2
Overview

• Best practices for building models that are not
  error-prone
• Testing and verifying models
• Verification tests whether the model does what
  you think it does
• e.g., it accurately solves the equations
• Validating models
• Validation tests whether the model is an
  appropriate representation of reality
• It solves the right equation
• Most of this is common sense

3
Best Practices for Building Models that are Not
Error-Prone

• Document your assumptions in the model as you
  build your model
• Notes, Text boxes
• Use descriptive names

4
(No Transcript)
5
Best Practices for Building Models that are Not
Error-Prone

• Document your assumptions in the model as you
  build your model
• Notes, Text boxes
• Use descriptive names
• Break complex calculations into separate elements
• Do no bury details in input fields

6
(No Transcript)
7
(No Transcript)
8
Best Practices for Building Models that are Not
Error-Prone

• Document your assumptions in the model as you
  build your model
• Notes, Text boxes
• Use descriptive names
• Break complex calculations into separate elements
• Do no bury details in input fields
• Use arrays (vectors or matrices) or cloned
  Containers for identical parallel calculations

9
Using Cloned Containers
Localized
Localized
A
B
A
B
Clone2
Clone1
X f(A,B)
X f(A,B)
10
Best Practices for Building Models that are Not
Error-Prone

• Dont put unit conversions into your model

11
(No Transcript)
12
Best Practices for Building Models that are Not
Error-Prone

• Dont put unit conversions into your model
• Dont use an equality test

13
(No Transcript)
14
Best Practices for Building Models that are Not
Error-Prone

• Dont put unit conversions into your model
• Dont use an equality test
• Any other suggestions or comments?

15
Testing and Verifying Models

• To the extent possible, break your model into
  modules (e.g., functions or subroutines) that can
  be testing separately
• Test over a range of variables
• Compare to independent calculations (spreadsheet,
  analytical solution)
• Use Interrupt elements to check for invalid
  behavior and bad logic

16
(No Transcript)
17
Testing and Verifying Models

• To the extent possible, break your model into
  modules (e.g., functions or subroutines) that can
  be testing separately
• Test over a range of variables
• Compare to independent calculations (spreadsheet,
  analytical solution)
• Use Interrupt elements to check for invalid
  behavior and bad logic
• Other ideas or suggestions?

18
Validating Models

• If possible, compare predicted results to
  historical observations
• Can use GoldSims Optimization features to
  calibrate predicted results to observed results
  (typically entered in a Time Series)

19
(No Transcript)
20
Final Words

• The time spent ensuring quality will be repaid
  many times over
• You must be thinking about this from the
  beginning of the project dont wait until the
  end