| Type: | Package |
| Title: | Neural Network with Levenberg-Marquardt Optimization |
| Version: | 1.0.1 |
| Description: | An implementation of a Neural Network using the Levenberg-Marquardt optimization from 'minpack.lm', ideal for small datasets. For more details see Moré (1978) <doi:10.1007/BFb0067700>. |
| License: | MIT + file LICENSE |
| BugReports: | https://github.com/umbe1987/nnetLM/issues |
| Imports: | minpack.lm, stats |
| Encoding: | UTF-8 |
| RoxygenNote: | 7.3.3 |
| URL: | https://github.com/umbe1987/nnetLM |
| NeedsCompilation: | no |
| Packaged: | 2026-02-20 11:00:06 UTC; minorum |
| Author: | Umberto Minora 
[aut, cre, cph] |
| Maintainer: | Umberto Minora <umbertofilippo@tiscali.it> |
| Repository: | CRAN |
| Date/Publication: | 2026-02-25 10:10:12 UTC |
Flattens the network parameters so they can be passed to [minipack.lm::nls.lm] 'par' argument
Description
Flattens the network parameters so they can be passed to [minipack.lm::nls.lm] 'par' argument
Usage
flatten_params(object)
Arguments
object |
an object of class "nnetLM" |
Value
flattened vector with network parameters (weights and biases)
Initalize the neural network object
Description
Initalize the neural network object
Usage
nnetLM(X, y, hidden, actFn)
Arguments
X |
Matrix of independent variables |
y |
Vector of dependent variables |
|
Vector of number of nodes in each hidden layer | |
actFn |
List of activation functions (must be length(hidden)+1 for the output node) |
Details
The activation functions within [actFn] list can be any existing or user-defined function. They must have a single numeric argument (e.g. [x]), and must return a numeric value of the same length as [x].
Value
An object with S3 class "nnetLM"
Examples
set.seed(123)
x <- seq(-10, 10, by = 0.1)
y <- sin(x) + rnorm(length(x), mean = 0, sd = 0.1)
X <- matrix(x, nrow = length(x), ncol = 1)
hidden <- c(10)
linear <- function(x) x
actFn <- list(tanh, linear)
nnet.obj <- nnetLM(X, y, hidden, actFn)
Performs a forward pass
Description
Performs a forward pass
Usage
## S3 method for class 'nnetLM'
predict(object, newdata)
Arguments
object |
a trained network object of class "nnetLM" |
newdata |
Matrix of predictors |
Value
a numeric vector with predicted values
Examples
set.seed(123)
x <- seq(-10, 10, by = 0.1)
y <- sin(x) + rnorm(length(x), mean = 0, sd = 0.1)
X <- matrix(x, nrow = length(x), ncol = 1)
hidden <- c(10)
linear <- function(x) x
actFn <- list(tanh, linear)
nnet.obj <- nnetLM(X, y, hidden, actFn)
nnet.obj <- train.nnetLM(nnet.obj,50)
pred.nnetLM <- predict(nnet.obj, X)
Residual function needed by [minipack.lm::nls.lm]
Description
Residual function needed by [minipack.lm::nls.lm]
Usage
residFun(params, observed, object, xx)
Arguments
params |
vector of flattened network parameters (weights and biases) |
observed |
an object of class "nnetLM" |
object |
an object of class "nnetLM" |
xx |
an object of class "nnetLM" |
Value
unflattened list with network parameters (weights and biases)
Train the neural network with Levenberg-Marquardt optimization using [minipack.lm::nls.lm]
Description
Train the neural network with Levenberg-Marquardt optimization using [minipack.lm::nls.lm]
Usage
train.nnetLM(object, epochs, progress = FALSE)
Arguments
object |
an object of class "nnetLM" |
epochs |
maximum number of iteration |
progress |
flag for printing network progress. Default is FALSE |
Value
the trained network object
See Also
[minipack.lm::nls.lm()]
Examples
x <- seq(-10, 10, by = 0.1)
y <- sin(x) + rnorm(length(x), mean = 0, sd = 0.1)
X <- matrix(x, nrow = length(x), ncol = 1)
hidden <- c(10)
linear <- function(x) x
actFn <- list(tanh, linear)
nnet.obj <- nnetLM(X, y, hidden, actFn)
nnet.obj <- train.nnetLM(nnet.obj,1)
Unflattens the network parameters after they have been used by [minipack.lm::nls.lm]
Description
Unflattens the network parameters after they have been used by [minipack.lm::nls.lm]
Usage
unflatten_params(params, object)
Arguments
params |
vector of flattened network parameters (weights and biases) |
object |
an object of class "nnetLM" |
Value
unflattened list with network parameters (weights and biases)