Setup hyperparameters for TabNet training.
Usage
setup_TabNet(
batch_size = 1024^2,
penalty = 0.001,
clip_value = NULL,
loss = "auto",
epochs = 50L,
drop_last = FALSE,
decision_width = NULL,
attention_width = NULL,
num_steps = 3L,
feature_reusage = 1.3,
mask_type = "sparsemax",
virtual_batch_size = 256^2,
valid_split = 0,
learn_rate = 0.02,
optimizer = "adam",
lr_scheduler = NULL,
lr_decay = 0.1,
step_size = 30,
checkpoint_epochs = 10L,
cat_emb_dim = 1L,
num_independent = 2L,
num_shared = 2L,
num_independent_decoder = 1L,
num_shared_decoder = 1L,
momentum = 0.02,
pretraining_ratio = 0.5,
device = "auto",
importance_sample_size = NULL,
early_stopping_monitor = "auto",
early_stopping_tolerance = 0,
early_stopping_patience = 0,
num_workers = 0L,
skip_importance = FALSE,
ifw = FALSE
)Arguments
- batch_size
(Tunable) Positive integer: Batch size.
- penalty
(Tunable) Numeric: Regularization penalty.
- clip_value
Numeric: Clip value.
- loss
Character: Loss function.
- epochs
(Tunable) Positive integer: Number of epochs.
- drop_last
Logical: If TRUE, drop last batch.
- decision_width
(Tunable) Positive integer: Decision width.
- attention_width
(Tunable) Positive integer: Attention width.
- num_steps
(Tunable) Positive integer: Number of steps.
- feature_reusage
(Tunable) Numeric: Feature reusage.
- mask_type
Character: Mask type.
- virtual_batch_size
(Tunable) Positive integer: Virtual batch size.
- valid_split
Numeric: Validation split.
- learn_rate
(Tunable) Numeric: Learning rate.
- optimizer
Character or torch function: Optimizer.
- lr_scheduler
Character or torch function: "step", "reduce_on_plateau".
- lr_decay
Numeric: Learning rate decay.
- step_size
Positive integer: Step size.
- checkpoint_epochs
(Tunable) Positive integer: Checkpoint epochs.
- cat_emb_dim
(Tunable) Positive integer: Categorical embedding dimension.
- num_independent
(Tunable) Positive integer: Number of independent Gated Linear Units (GLU) at each step of the encoder.
(Tunable) Positive integer: Number of shared Gated Linear Units (GLU) at each step of the encoder.
- num_independent_decoder
(Tunable) Positive integer: Number of independent GLU layers for pretraining.
(Tunable) Positive integer: Number of shared GLU layers for pretraining.
- momentum
(Tunable) Numeric: Momentum.
- pretraining_ratio
(Tunable) Numeric: Pretraining ratio.
- device
Character: Device "cpu" or "cuda".
- importance_sample_size
Positive integer: Importance sample size.
- early_stopping_monitor
Character: Early stopping monitor. "valid_loss", "train_loss", "auto".
- early_stopping_tolerance
Numeric: Minimum relative improvement to reset the patience counter.
- early_stopping_patience
Positive integer: Number of epochs without improving before stopping.
- num_workers
Positive integer: Number of subprocesses for data loacding.
- skip_importance
Logical: If TRUE, skip importance calculation.
- ifw
Logical: If TRUE, use Inverse Frequency Weighting in classification.