Fit a hydromad model by sampling the parameter space.

Fit a hydromad model by sampling the parameter space. Returns best result from sampling in parameter ranges using random, latin hypercube sampling, or a uniform grid (all combinations). The function also retains the parameter sets and objective function values, which can be used to define a feasible parameter set

fitBySampling(
  MODEL,
  objective = hydromad.getOption("objective"),
  samples = hydromad.getOption("fit.samples"),
  sampletype = c("latin.hypercube", "random", "all.combinations")
)

Arguments

MODEL: a model specification created by hydromad. It should not be fully specified, i.e one or more parameters should be defined by ranges of values rather than exact values.
objective: objective function to maximise, given as a function(Q, X, ...). See objFunVal.
samples: number of parameter sets to test.
sampletype: sampling scheme -- see parameterSets.

Value

the best model from those sampled, according to the given objective function.

Details

See parameterSets.

Author

Felix Andrews felix@nfrac.org

Examples



data(Cotter)
x <- Cotter[1:1000]

## IHACRES CWI model with armax unit hydrograph fitted by least squares
modx <- hydromad(x, sma = "cwi", routing = "armax", rfit = "ls")
modx
#> 
#> Hydromad model with "cwi" SMA and "armax" routing:
#> Start = 1966-05-01, End = 1969-01-24
#> 
#> SMA Parameters:
#>       lower upper     
#> tw        0   100     
#> f         0     8     
#> scale    NA    NA     
#> l         0     0 (==)
#> p         1     1 (==)
#> t_ref    20    20 (==)
#> Routing Parameters:
#> NULL
#> Routing fit spec.: "ls" 

foo <- fitBySampling(modx)

summary(foo)
#> 
#> Call:
#> hydromad(DATA = x, sma = "cwi", routing = "armax", rfit = "ls", 
#>     tw = 95.9596, f = 3.87879, l = 0, p = 1, t_ref = 20, scale = 0.00134404)
#> 
#> Time steps: 900 (0 missing).
#> Runoff ratio (Q/P): (0.7028 / 2.285) = 0.3075
#> rel bias: 0.06098
#> r squared: 0.7004
#> r sq sqrt: 0.8015
#> r sq log: 0.814
#> 
#> For definitions see ?hydromad.stats
#> 

## plot objective function value improvement over time
xyplot(optimtrace(foo),
  type = "b",
  xlab = "function evaluations", ylab = "objective fn. value"
)