model.adapter-class: Abstraction layer for model functions/objects
In Marchen/model.adapter: Abstraction Layer for Statistical/Machine Learning Models
model.adapter-class R Documentation
Abstraction layer for model functions/objects
Description
This class encapsulates differences in specifications of
statistical/machine learning functions and model objects to provide a
standardized way to access data and common methods of models.
To add support for a modeling function, new generator object of a
model.interface class must be implimented and methods that
cannot work well with the model should be overriden.
Methods
model.adapter$new(x, envir = parent.frame(2L), data = NULL, package.name = NULL)
Generate a new object.
Args
x-
an object of supported models or a call for a model
function.
envir = parent.frame(2L)-
an environment in which call in x is evaluated.
Default value is the environment where the object of
model.adapter was created.
data = NULL-
a data.frame used for further manipulations.
package.name = NULL-
a character specifying package name of the
function/object. If not specified, package.name is
automatically determined from x. This is mainly
used to control functions with same name in different
packages (e.g., gam in gam and
mgcv packages).
model.adapter$x.names(specials = NULL, type = c("all", "base"))
Access names of explanatory variables.
Args
specials = NULL-
special characters to be passed to terms.
type = c("all", "base")-
if "all", this function returns all explanatory variables
including interactions, higher order terms, splines, etc.
If "base" only basic form of the variables are returned.
Returns
a character vector of names of explanatory variables.
model.adapter$y.names()
Access names of response variables.
model.adapter$predict(newdata = NULL, type = c("response", "link", "prob", "class"), random = ~0, ...)
Call predict method.
For regression models, offset term is adjusted if 1) type is not "link",
2) the model having only one offset term, 3) offset was specified as
offset = offset or formula = y ~ x + offset(offset)
rather than offset = data$offset or
formula = y ~ x + offset(data$offset), 4) the model is not
glmmML or ranger.
Also, it is users responsibility to use offset term with appropreate
scale. For example if the model is glm with poisson
family with log link function, the offset term should be
offset = log(offset) or
formula = y ~ x + offset(log(offset)).
Args
newdata = NULL-
a data.frame containing data used for
prediction.
-
type = c("response", "link", "prob", "class")
-
the type of prediciton. "response" and "link" can be
used for generalized linear (mixed) model and spcify
scale of prediction.
"response" is on the scale of the response variable.
"link" is on the scale of the link function.
"prob" and "class" are used for classification models.
"prob" calculate probability of being each class of
response variable.
"class" makes predicted class for each observation.
random = ~0-
the random effect to use.
Tentatively, ~0 means don't use random effects.
...-
other variables passed to predict methods.
Returns
a
model.adapter.prediction
object.
model.adapter$residuals(type = c("response", "link"))
Calculate residuals of the model.
Args
- type = c("response", "link")
-
a character specifying type of residual (i.e., scale of
residual) to be calculated.
Returns
a numeric vector of residuals.
Details
When calling model.adapter$new using model call by
do.call, please specify quote = TRUE.
Without it, a call of function in x is evaluated and unintended
result may be returned.
Active bindings
calla read-only call object used for initialization of
model.adapter or a call object which used for
construction of the model object.
Note that call in this field is modified to have their full
names by match.call() function.
Therefore, the call in this field doesn't need to be identical
to original call used for initialization of the class.
When the original call used for creating the model object could
not be recovered from the object, this field becomes NULL.
objecta read-only object of the model.
If model.adapter is initialized by a call for a model
function, this field returns a object created by evaluating the
call.
dataa read-only data.frame used for modeling.
At least, all columns used for the modeling are stored in this field.
Because some modeling function doesn't keep original data.frame
used for modeling or call in resultant object, this field can't
have the same data.frame used for modeling in such situation.
When all the columns used for the modeling is not available from this
field becomes NULL.
Also, when the model doesn't store original data and some function
such as 'log()' was applied in model construction, the 'data.frame'
stored in this field can be different from original values.
At least, this is applicable for 'glmer' and 'lmer'.
formulaa read-only formula object specifying structure of the model.
'.' in the formula object is expanded so that this field may not
be same as the original formula specified in the call for
the model function or the model object in such case.
familya read-only character of family name.
If the model does not use family, this field is NULL.
linka function object of link function of the model. If the model does
not have link function, this field has identity
function.
linkinva function object of inverse link function of the model. If the model
does not have inverse link function, this field has
identity function.
package.namea read-only character spcifying package name of the model.
model.typea character representing model type. The possible values are
"regression" and "classification".
x.varsa data.frame having all explanatory variables.
y.varsa data.frame having all response variables.
Methods
Public methods
-
-
-
-
-
-
Method new()
Usage
model.adapter$new(
x,
envir = parent.frame(2L),
data = NULL,
package.name = NULL
)
Method x.names()
Usage
model.adapter$x.names(specials = NULL, type = c("all", "base"))
Method y.names()
Usage
model.adapter$y.names()
Method predict()
Usage
model.adapter$predict(
newdata = NULL,
type = c("response", "link", "prob", "class"),
random = ~0,
...
)
Method residuals()
Usage
model.adapter$residuals(type = c("response", "link"))
Method clone()
The objects of this class are cloneable with this method.
Usage
model.adapter$clone(deep = FALSE)
Arguments
deepWhether to make a deep clone.
Marchen/model.adapter documentation built on July 3, 2023, 11:07 a.m.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
| model.adapter-class | R Documentation |
Abstraction layer for model functions/objects
Description
This class encapsulates differences in specifications of
statistical/machine learning functions and model objects to provide a
standardized way to access data and common methods of models.
To add support for a modeling function, new generator object of a
model.interface class must be implimented and methods that
cannot work well with the model should be overriden.
Methods
model.adapter$new(x, envir = parent.frame(2L), data = NULL, package.name = NULL)
Generate a new object.
Args
x-
an object of supported models or a call for a model function.
envir = parent.frame(2L)-
an environment in which call in x is evaluated. Default value is the environment where the object of model.adapter was created.
data = NULL-
a data.frame used for further manipulations.
package.name = NULL-
a character specifying package name of the function/object. If not specified, package.name is automatically determined from
x. This is mainly used to control functions with same name in different packages (e.g.,gamingamandmgcvpackages).
model.adapter$x.names(specials = NULL, type = c("all", "base"))
Access names of explanatory variables.
Args
specials = NULL-
special characters to be passed to terms.
type = c("all", "base")-
if "all", this function returns all explanatory variables including interactions, higher order terms, splines, etc. If "base" only basic form of the variables are returned.
Returns
a character vector of names of explanatory variables.
model.adapter$y.names()
Access names of response variables.
model.adapter$predict(newdata = NULL, type = c("response", "link", "prob", "class"), random = ~0, ...)
Call predict method.
For regression models, offset term is adjusted if 1) type is not "link",
2) the model having only one offset term, 3) offset was specified as
offset = offset or formula = y ~ x + offset(offset)
rather than offset = data$offset or
formula = y ~ x + offset(data$offset), 4) the model is not
glmmML or ranger.
Also, it is users responsibility to use offset term with appropreate
scale. For example if the model is glm with poisson
family with log link function, the offset term should be
offset = log(offset) or
formula = y ~ x + offset(log(offset)).
Args
newdata = NULL-
a
data.framecontaining data used for prediction. -
type = c("response", "link", "prob", "class") -
the type of prediciton. "response" and "link" can be used for generalized linear (mixed) model and spcify scale of prediction. "response" is on the scale of the response variable. "link" is on the scale of the link function. "prob" and "class" are used for classification models. "prob" calculate probability of being each class of response variable. "class" makes predicted class for each observation.
random = ~0-
the random effect to use. Tentatively, ~0 means don't use random effects.
...-
other variables passed to predict methods.
Returns
a
model.adapter.prediction
object.
model.adapter$residuals(type = c("response", "link"))
Calculate residuals of the model.
Args
- type = c("response", "link")
-
a character specifying type of residual (i.e., scale of residual) to be calculated.
Returns
a numeric vector of residuals.
Details
When calling model.adapter$new using model call by
do.call, please specify quote = TRUE.
Without it, a call of function in x is evaluated and unintended
result may be returned.
Active bindings
calla read-only
callobject used for initialization ofmodel.adapteror acallobject which used for construction of the model object. Note thatcallin this field is modified to have their full names bymatch.call()function. Therefore, thecallin this field doesn't need to be identical to originalcallused for initialization of the class. When the originalcallused for creating the model object could not be recovered from the object, this field becomesNULL.objecta read-only object of the model. If
model.adapteris initialized by acallfor a model function, this field returns a object created by evaluating thecall.dataa read-only
data.frameused for modeling. At least, all columns used for the modeling are stored in this field. Because some modeling function doesn't keep originaldata.frameused for modeling orcallin resultant object, this field can't have the samedata.frameused for modeling in such situation. When all the columns used for the modeling is not available from this field becomesNULL. Also, when the model doesn't store original data and some function such as 'log()' was applied in model construction, the 'data.frame' stored in this field can be different from original values. At least, this is applicable for 'glmer' and 'lmer'.formulaa read-only
formulaobject specifying structure of the model. '.' in theformulaobject is expanded so that this field may not be same as the originalformulaspecified in thecallfor the model function or the model object in such case.familya read-only character of family name. If the model does not use family, this field is
NULL.linka function object of link function of the model. If the model does not have link function, this field has
identityfunction.linkinva function object of inverse link function of the model. If the model does not have inverse link function, this field has
identityfunction.package.namea read-only character spcifying package name of the model.
model.typea character representing model type. The possible values are "regression" and "classification".
x.varsa data.frame having all explanatory variables.
y.varsa data.frame having all response variables.
Methods
Public methods
Method new()
Usage
model.adapter$new( x, envir = parent.frame(2L), data = NULL, package.name = NULL )
Method x.names()
Usage
model.adapter$x.names(specials = NULL, type = c("all", "base"))Method y.names()
Usage
model.adapter$y.names()
Method predict()
Usage
model.adapter$predict(
newdata = NULL,
type = c("response", "link", "prob", "class"),
random = ~0,
...
)Method residuals()
Usage
model.adapter$residuals(type = c("response", "link"))Method clone()
The objects of this class are cloneable with this method.
Usage
model.adapter$clone(deep = FALSE)
Arguments
deepWhether to make a deep clone.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Embedding an R snippet on your website
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.