creates settings for the Estimator, which takes a model and trains it
Usage
setEstimator(
learningRate = "auto",
weightDecay = 0,
batchSize = 512,
epochs = 30,
device = "cpu",
optimizer = torch$optim$AdamW,
scheduler = list(fun = torch$optim$lr_scheduler$ReduceLROnPlateau, params =
list(patience = 1)),
criterion = torch$nn$BCEWithLogitsLoss,
earlyStopping = list(useEarlyStopping = TRUE, params = list(patience = 4)),
compile = FALSE,
metric = "auc",
accumulationSteps = NULL,
seed = NULL,
trainValidationSplit = FALSE
)Arguments
- learningRate
what learning rate to use
- weightDecay
what weight_decay to use
- batchSize
batchSize to use
- epochs
how many epochs to train for
- device
what device to train on, can be a string or a function to that evaluates to the device during runtime
- optimizer
which optimizer to use
- scheduler
which learning rate scheduler to use
- criterion
loss function to use
- earlyStopping
If earlyStopping should be used which stops the training of your metric is not improving
- compile
if the model should be compiled before training, default FALSE
- metric
either `auc` or `loss` or a custom metric to use. This is the metric used for scheduler and earlyStopping. Needs to be a list with function `fun`, mode either `min` or `max` and a `name`, `fun` needs to be a function that takes in prediction and labels and outputs a score.
- accumulationSteps
how many steps to accumulate gradients before updating weights, can also be a function that is evaluated during runtime
- seed
seed to initialize weights of model with
- trainValidationSplit
if TRUE, perform a train-validation split for model selection instead of cross validation