Transcript Slide 1

RobOff software and GUI
1. GUI basics
2. Flow of use
3. Optimization
4. Summary of inputs and outputs
5. Exercises
Teaching material
• Software as installer executable or zip package
RobOff GUI
Setup
Results
Opt.
Log
Tabs for different entities
Sections of the GUI
• Setup: load / edit / save RobOff setups
– Environments, features, actions, responses and all
other parameters
• Results: calculate / save / visualize
– Summary and plots for different dimensions:
features, environments, actions, etc.
• Optimization: calculate / load / save optimal
allocations
• Several examples included with the software: just
load, edit and run
Input: two alternatives
Plain text files
GUI tabs
Run: two alternatives
Command line (automated/scripting)
GUI (results + opt)
Analyze results: two alternatives
Output (plain) text files
GUI (results + opt)
Flow of use
Setup
section
Results +
opt.
sections
Optimization
• Maximize efficiency (return on investment)
– Budget (B)
– Budget resolution
• Actions:
– Costs (constant, function of area, time-dependent)
– Area availability
• How do you allocate B among actions?
• Efficiency measure: one of the variants of
sustainability / conservation value
• Robustness requirement: robust / opportunity
Optimization methods
Method
Effective max
Computation speed
problem size (approx.)
Greedy search
Complexi Optimality
ty to user
>= Hundreds of
Extremely fast, in
action-environment
seconds
low
Potentially very
suboptimal.
pairs (or more)
Grid-based
8-10 dimensions
Fast with small dimensions
exhaustive
(action-environment
but deteriorates
search
pairs) within 1 hour
low
Up to resolution
limited by
dimensionality
exponentially with the
problem dimension
Exh. search +
As for exhaustive
As exhaustive search, but
local search
search
an order of magnitude
low
Globally optimal for
convex problems
slower.
Stochastic
Tens of action-
Computation time can be
global search
environment pairs
limited. Several repetitions
within 1 hour.
are recommended to verify
convergence.
medium
Not guaranteed but the
only practical option
for large non-convex
problems.
Levels or groups of actions
• Mandatory
– set a priori in the “allocations” tab
– always enforced (even when optimizing)
• Preset
– convenient to define different allocation scenarios
• Optimized
– solution(s) found by RobOff, different criteria:
• strong/weak across environments or features
• robust / nominal / opportunity
– Save/load from the “optimization” section
Summary of input files
Summary of output files
Text files
GUI
Exercise and Q & A
• Load one of the example setups and familiarize yourself
with the tabs of the Setup and Results sections
• If you were starting a setup from scratch, what would you
define first?
• Ways to obtain or derive responses?
• How to select the type/shape of the benefit functions?
• What about the y-axis of the responses (occurrence levels)
and the x-axis of the benefit functions? Do they need to
match?
• Environments have these two parameters: weight and
condition. Aren’t they redundant?
• How do you change the spatial scope of analysis
(assuming same features, actions and responses).