quantV {qualV}R Documentation

Quantitative Validation Methods

Description

Different methods for calculating the difference between two vectors.

Usage

generalME(o, p,
          ignore   = c("raw", "centered", "scaled", "ordered"),
          geometry = c("real", "logarithmic", "geometric", "ordinal"),
          measure  = c("mad", "var", "sd"),
          type     = c("dissimilarity", "normalized", "similarity",
                       "reference", "formula", "name", "function"),
                       method = NULL)
   MAE(o, p, type = "dissimilarity")
  MAPE(o, p, type = "dissimilarity")
   MSE(o, p, type = "dissimilarity")

  RMSE(o, p, type = "dissimilarity")
  CMAE(o, p, type = "dissimilarity")
  CMSE(o, p, type = "dissimilarity")
 RCMSE(o, p, type = "dissimilarity")
  SMAE(o, p, type = "dissimilarity")
  SMSE(o, p, type = "dissimilarity")
 RSMSE(o, p, type = "dissimilarity")
  MALE(o, p, type = "dissimilarity")
  MAGE(o, p, type = "dissimilarity")
 RMSLE(o, p, type = "dissimilarity")
 RMSGE(o, p, type = "dissimilarity")

 SMALE(o, p, type = "dissimilarity")
 SMAGE(o, p, type = "dissimilarity")
 SMSLE(o, p, type = "dissimilarity")

RSMSLE(o, p, type = "dissimilarity")
RSMSGE(o, p, type = "dissimilarity")

  MAOE(o, p, type = "dissimilarity")
  MSOE(o, p, type = "dissimilarity")
 RMSOE(o, p, type = "dissimilarity")

Arguments

o vector of observed values
p vector of corresponding predicted values
type one of "dissimilarity", "normalized", "similarity", "reference", "formula", for the dissimilarity measure, the normalized dissimilarity measure, the similarity measure, or the formula for the normalized measure. For generalME it is additionally possible to specify "function" for getting the corresponding function and "name" for getting the name of the function.
ignore specifies which aspects should be ignored: "raw" compares original values, "centered" removes differences in mean, "scaled" ignores scaling, "ordered" indicates the use of the ordinal geometry only.
geometry indicating the geometry to be used for the data and the output, "real" corresponds to arithmetic differences and means, "logarithmic" to handling relative data on a logarithmic scale, "geometric" to geometric means and differences and "ordinal" to a pure ordinal treatment.
measure indicates how distances should be measured: as mean absolute distances like in MAD, as squared distances like in a variance, or as the root of mean squared distances like in sd.
method optionally the function to be used can specified directly as a function or as a string.

Details

These comparison criteria are designed for a semiquantitative comparison of observed values o with predicted values p to validate the performance of the prediction.
The general naming convention follows the grammar scheme
[R][C|S]M[S|A][L|G|O]E
corresponding to [Root] [Centered | Scaled] Mean [Squared | Absolute]
[Logarithmic, Geometric, Ordinal] Error

Root
is used together with squared errors to indicate, that a root is applied to the mean.
Centered
indicates that an additive constant is allowed.
Scaled
indicates that a scaling of the predictive sequence is allowed. Scaled implies centered for real scale.
Squared
indicates that squared error is used.
Absolute
indicates that absolute error is used.
Logarithmic
indicates that the error is calculated based on the logarithms of the values. This is useful for data on a relative scale.
Geometric
indicates that the result is to be understood as a factor, similar to a geometric mean.
Ordinal
indicates that only the order of the observations is taken into account by analyzing the data by ranks scaled to the interval [0, 1].

The mean errors for squared error measures are based on the number of degrees of freedom of the residuals.

Value

generalME selects the best deviance measure according to the description given in the parameters. It has the two additional possibilities of name and function in the type parameter.
MAE mean absolute error 1/n
MAPE mean absolute percentage error
MSE mean squared error
RMSE root mean squared error
CMAE centered mean absolute error
CMSE centered mean squared error
RCMSE root centered mean squared error
SMAE scaled mean absolute error
SMSE scaled mean squared error
RSMSE root scaled mean squared error
MALE mean absolute logarithmic error
MAGE mean absolute geometric error
MSLE mean squared logarithmic error
MSGE mean squared geometric error
RMSLE root mean squared logarithmic error
SMALE scaled mean absolute logarithmic error
SMAGE scaled mean absolute relative error
SMSLE scaled mean squared logarithmic error
RSMSLE root scaled mean squared logarithmic error
RSMSGE root scaled mean squared geometric error
MAOE mean absolute ordinal error
MSOE mean squared ordinal error
RMSOE root mean squared ordinal error

References

Mayer, D. G. and Butler, D. G. (1993) Statistical Validation. Ecological Modelling, 68, 21-32.

Jachner, S., K.G. v.d. Boogaart, T. Petzoldt (2007) Statistical methods for the qualitative assessment of dynamic models with time delay (R package qualV), in preparation

See Also

EF, GRI, compareME

Examples

data(phyto)
obsb <- na.omit(obs[match(sim$t, obs$t), ])
simb <- sim[na.omit(match(obs$t, sim$t)), ]
o <- obsb$y
p <- simb$y

generalME(o, p, ignore = "raw", geometry = "real")

   MAE(o, p)
  MAPE(o, p)
   MSE(o, p)
  RMSE(o, p)
  CMAE(o, p)
  CMSE(o, p)
 RCMSE(o, p)
  SMAE(o, p)
  SMSE(o, p)
 RSMSE(o, p)
  MALE(o, p)
  MAGE(o, p)
 RMSLE(o, p)
 RMSGE(o, p)

 SMALE(o, p)
 SMAGE(o, p)
 SMSLE(o, p)

RSMSLE(o, p)
RSMSGE(o, p)

  MAOE(o, p)
  MSOE(o, p)
 RMSOE(o, p)
   MAE(o, p)
  MAPE(o, p)

   MSE(o, p, type = "s")
  RMSE(o, p, type = "s")
  CMAE(o, p, type = "s")
  CMSE(o, p, type = "s")
 RCMSE(o, p, type = "s")
  SMAE(o, p, type = "s")
  SMSE(o, p, type = "s")
 RSMSE(o, p, type = "s")
  MALE(o, p, type = "s")
  MAGE(o, p, type = "s")
 RMSLE(o, p, type = "s")
 RMSGE(o, p, type = "s")

 SMALE(o, p, type = "s")
 SMAGE(o, p, type = "s")
 SMSLE(o, p, type = "s")

RSMSLE(o, p, type = "s")
RSMSGE(o, p, type = "s")

  MAOE(o, p, type = "s")
  MSOE(o, p, type = "s")
 RMSOE(o, p, type = "s")

[Package qualV version 0.2-4 Index]