Diagnostics with gadget

1
Diagnostics with gadget

• Morten Nygaard Åsnes

2

• The fleksibest Model
• Complex model with
• 133 parameters to optimise
• 72 time steps, 10 age groups, 120 length groups
• 2 sub stocks, 2 modelled fleets
• 11 likelihood components

3

• Diagnostics are needed when
• Setting up a new model
• Changing or extending a model
• Running a model

4

• Setting up a model
• Model consistency
• Choosing model formulations
• Data consistency
• Weighting different data sources
• Parameters

5

• Running the model
• Did it optimise to a solution?
• Did it optimise to the global solution?
• Is the solution sensible?
• How does it compare to earlier runs, or to
  other models?

6

• Model output
• Parameter values and likelihood score (both
  final, and during optimisation)
• Standard output of model stocks (text files)
• Output from the optimiser
• Must post process to make use of this

7

• Analysing the optimisation
• Sensitivity plots of the likelihood
• Sensitivity plots of individual components
• Done by sampling the likelihood function again
  (slow)
• Often enough to only sample a few points.

8
(No Transcript)
9
(No Transcript)
10
(No Transcript)
11
(No Transcript)
12

• How well does the model fit to data?
• Comparing model with observations
• Plotting the likelihood contributions
• How well does individual components fit with
  data?
• Can help finding errors in data (outliers)

13
(No Transcript)
14
(No Transcript)
15
(No Transcript)
16
(No Transcript)
17
(No Transcript)
18
(No Transcript)
19

• Model results
• Textual output
• Plots of population development
• Verify that the solution is sensible,
• Compare against other runs, or other models.
• Plot retrospective patterns

20
(No Transcript)
21
(No Transcript)
22
(No Transcript)
23
(No Transcript)
24
(No Transcript)
25
(No Transcript)
26
(No Transcript)
27
(No Transcript)