#Dave Woods, University of Southampton, UK, 2005
#email: dcw@maths.soton.ac.uk
#WWW: http://www.maths.soton.ac.uk/staff/woods

This readme file briefly describes the input for the "perturbations.cpp" assessment algorithm
for compromise designs under GLMs. The file "link.in" has the input for the same
example as this readme and will work with the algorithm.

Running the code
After compliation, run the executable. The three commands are read from the default input and
give the name of the main input file (link.in in this example), the main output file (link.
out) and the design output file (link_des.out; these are listed in link_input.in). The input file
specifies the parameters for the simulated annealing algorithm, the models to be 
considered, and the parameter space of interest (see comments below).

Important: to change the random number seed, run the executable (coeff.run) with the
commandline

GSL_RNG_SEED=12 pert.out

where "12" sould be replaced with your chosen seed. (The seed is not set dynamically, e.g.
using the current time, to aid parallel execution on multi=processor machines).

The output from the algorithm is a table of objective function values for each set of 
models (parameter vectors) and designs. These can then be imported into your favourite 
spreadsheet program and manipulated.

Below is an annotated version of "pert.in"


1000 #number of simulations

#annealing parameters - see the Simulated Annealing algorithm input readme
1000
0.95
50
10
0.75
200
0.0001

2 #number of variables
6 #number of runs

2 #number of different link functions

#number of linear predictors for each link function
2
2

#number of parameters in each linear predictor
3 4
3 4

#model for each linear predictor, specified as a matrix - see the Simulated Annealing 
#algorithm input readme
0 0
1 0
0 1

0 0
1 0
0 1
1 1

0 0
1 0
0 1

0 0
1 0
0 1
1 1

#specify the link functions
logit
comp

#objective function weights for each model
1
1
1
1

#perburbation parameters for each set of models within a link function
#link 1, model 2 (smallest model)
0.3
0.3
0.3

#link 1, model 1 (largest model - therefore no perturbation)
0
0
0
0

#link 2, model 2 (smallest model)
0.3
0.3
0.3

#link 2, model 1 (largest model - therefore no perturbation)
0
0
0
0

#ranges for each variable
-1 1
-1 1

#parameters for the normal perturbation distribution
#means
0
0
0
0

#var-cov matrix (currently, covariances are ignored)
0 0 0 0
0 6 0 0
0 0 6 0
0 0 0 6
