This wraps a torch::nn_loss and annotates it with metadata, most importantly a ParamSet.
The loss function is created for the given parameter values by calling the $generate() method.
This class is usually used to configure the loss function of a torch learner, e.g.
when construcing a learner or in a ModelDescriptor.
For a list of available losses, see mlr3torch_losses.
Items from this dictionary can be retrieved using t_loss().
Parameters
Defined by the constructor argument param_set.
If no parameter set is provided during construction, the parameter set is constructed by creating a parameter
for each argument of the wrapped loss function, where the parameters are then of type ParamUty.
See also
Other Torch Descriptor:
TorchCallback,
TorchDescriptor,
TorchOptimizer,
as_torch_callbacks(),
as_torch_loss(),
as_torch_optimizer(),
mlr3torch_losses,
mlr3torch_optimizers,
t_clbk(),
t_loss(),
t_opt()
Super class
TorchDescriptor -> TorchLoss
Public fields
task_types(
character())
The task types this loss supports.
Methods
Inherited methods
TorchLoss$new()
Creates a new instance of this R6 class.
Usage
TorchLoss$new(
torch_loss,
task_types = NULL,
param_set = NULL,
id = NULL,
label = NULL,
packages = NULL,
man = NULL
)Arguments
torch_loss(
nn_lossorfunction)
The loss module or function that generates the loss module. Can have argumentstaskthat will be provided when the loss is instantiated.task_types(
character())
The task types supported by this loss. IfNULL(default), the loss is applicable to"classif","regr"and"torch".param_set(
ParamSetorNULL)
The parameter set. IfNULL(default) it is inferred fromtorch_loss.id(
character(1))
The id for of the new object.label(
character(1))
Label for the new instance.packages(
character())
The R packages this object depends on.man(
character(1))
String in the format[pkg]::[topic]pointing to a manual page for this object. The referenced help package can be opened via method$help().
Examples
# Create a new torch loss
torch_loss = TorchLoss$new(torch_loss = nn_mse_loss, task_types = "regr")
torch_loss
#> <TorchLoss:nn_mse_loss> nn_mse_loss
#> * Generator: nn_mse_loss
#> * Parameters: list()
#> * Packages: torch,mlr3torch
#> * Task Types: regr
# the parameters are inferred
torch_loss$param_set
#> <ParamSet(1)>
#> id class lower upper nlevels default value
#> <char> <char> <num> <num> <num> <list> <list>
#> 1: reduction ParamUty NA NA Inf <NoDefault[0]> [NULL]
# Retrieve a loss from the dictionary:
torch_loss = t_loss("mse", reduction = "mean")
# is the same as
torch_loss
#> <TorchLoss:mse> Mean Squared Error
#> * Generator: nn_mse_loss
#> * Parameters: reduction=mean
#> * Packages: torch,mlr3torch
#> * Task Types: regr,torch
torch_loss$param_set
#> <ParamSet(1)>
#> id class lower upper nlevels default value
#> <char> <char> <num> <num> <num> <list> <list>
#> 1: reduction ParamFct NA NA 2 mean mean
torch_loss$label
#> [1] "Mean Squared Error"
torch_loss$task_types
#> [1] "regr" "torch"
torch_loss$id
#> [1] "mse"
# Create the loss function
loss_fn = torch_loss$generate()
loss_fn
#> An `nn_module` containing 0 parameters.
# Is the same as
nn_mse_loss(reduction = "mean")
#> An `nn_module` containing 0 parameters.
# open the help page of the wrapped loss function
# torch_loss$help()
# Use in a learner
learner = lrn("regr.mlp", loss = t_loss("mse"))
# The parameters of the loss are added to the learner's parameter set
learner$param_set
#> <ParamSetCollection(40)>
#> id class lower upper nlevels default
#> <char> <char> <num> <num> <num> <list>
#> 1: epochs ParamInt 0e+00 Inf Inf <NoDefault[0]>
#> 2: device ParamFct NA NA 12 <NoDefault[0]>
#> 3: num_threads ParamInt 1e+00 Inf Inf <NoDefault[0]>
#> 4: num_interop_threads ParamInt 1e+00 Inf Inf <NoDefault[0]>
#> 5: seed ParamInt -Inf Inf Inf <NoDefault[0]>
#> 6: resume ParamUty NA NA Inf [NULL]
#> 7: eval_freq ParamInt 1e+00 Inf Inf <NoDefault[0]>
#> 8: measures_train ParamUty NA NA Inf <NoDefault[0]>
#> 9: measures_valid ParamUty NA NA Inf <NoDefault[0]>
#> 10: patience ParamInt 0e+00 Inf Inf <NoDefault[0]>
#> 11: min_delta ParamDbl 0e+00 Inf Inf <NoDefault[0]>
#> 12: restore_best_weights ParamLgl NA NA 2 <NoDefault[0]>
#> 13: batch_size ParamInt 1e+00 Inf Inf <NoDefault[0]>
#> 14: batch_size_predict ParamInt 1e+00 Inf Inf <NoDefault[0]>
#> 15: shuffle ParamLgl NA NA 2 FALSE
#> 16: sampler ParamUty NA NA Inf <NoDefault[0]>
#> 17: batch_sampler ParamUty NA NA Inf <NoDefault[0]>
#> 18: num_workers ParamInt 0e+00 Inf Inf 0
#> 19: collate_fn ParamUty NA NA Inf [NULL]
#> 20: pin_memory ParamLgl NA NA 2 FALSE
#> 21: drop_last ParamLgl NA NA 2 FALSE
#> 22: timeout ParamDbl -Inf Inf Inf -1
#> 23: worker_init_fn ParamUty NA NA Inf <NoDefault[0]>
#> 24: worker_globals ParamUty NA NA Inf <NoDefault[0]>
#> 25: worker_packages ParamUty NA NA Inf <NoDefault[0]>
#> 26: tensor_dataset ParamFct NA NA 1 <NoDefault[0]>
#> 27: jit_trace ParamLgl NA NA 2 <NoDefault[0]>
#> 28: neurons ParamUty NA NA Inf <NoDefault[0]>
#> 29: p ParamDbl 0e+00 1 Inf <NoDefault[0]>
#> 30: n_layers ParamInt 1e+00 Inf Inf <NoDefault[0]>
#> 31: activation ParamUty NA NA Inf <NoDefault[0]>
#> 32: activation_args ParamUty NA NA Inf <NoDefault[0]>
#> 33: shape ParamUty NA NA Inf <NoDefault[0]>
#> 34: opt.lr ParamDbl 0e+00 Inf Inf 0.001
#> 35: opt.betas ParamUty NA NA Inf 0.900,0.999
#> 36: opt.eps ParamDbl 1e-16 Inf Inf 1e-08
#> 37: opt.weight_decay ParamDbl 0e+00 Inf Inf 0
#> 38: opt.amsgrad ParamLgl NA NA 2 FALSE
#> 39: opt.param_groups ParamUty NA NA Inf <NoDefault[0]>
#> 40: loss.reduction ParamFct NA NA 2 mean
#> id class lower upper nlevels default
#> <char> <char> <num> <num> <num> <list>
#> value
#> <list>
#> 1: [NULL]
#> 2: auto
#> 3: 1
#> 4: [NULL]
#> 5: random
#> 6: [NULL]
#> 7: 1
#> 8: <list[0]>
#> 9: <list[0]>
#> 10: 0
#> 11: 0
#> 12: FALSE
#> 13: [NULL]
#> 14: [NULL]
#> 15: TRUE
#> 16: [NULL]
#> 17: [NULL]
#> 18: [NULL]
#> 19: [NULL]
#> 20: [NULL]
#> 21: [NULL]
#> 22: [NULL]
#> 23: [NULL]
#> 24: [NULL]
#> 25: [NULL]
#> 26: FALSE
#> 27: FALSE
#> 28:
#> 29: 0.1
#> 30: [NULL]
#> 31: <nn_relu[1]>
#> 32: <list[0]>
#> 33: [NULL]
#> 34: [NULL]
#> 35: [NULL]
#> 36: [NULL]
#> 37: [NULL]
#> 38: [NULL]
#> 39: [NULL]
#> 40: [NULL]
#> value
#> <list>