Recurrent QRN fitting function — QRN_seq

Used to fit a recurrent quantile regression neural network on a data sample. Use the QRN_fit_multiple() wrapper instead, with data_type="seq", for better stability using fitting restart.

Usage

QRN_seq_fit(
  X,
  Y,
  q_level,
  hidden_size = 10,
  num_layers = 1,
  rnn_type = c("lstm", "gru"),
  p_drop = 0,
  learning_rate = 1e-04,
  L2_pen = 0,
  seq_len = 10,
  scale_features = TRUE,
  n_epochs = 10000,
  batch_size = 256,
  X_valid = NULL,
  Y_valid = NULL,
  lr_decay = 1,
  patience_decay = n_epochs,
  min_lr = 0,
  patience_stop = n_epochs,
  tol = 1e-04,
  fold_separation = NULL,
  warm_start_path = NULL,
  patience_lag = 5,
  optim_met = "adam",
  seed = NULL,
  verbose = 2,
  device = default_device()
)

Arguments

X: Matrix of covariates, for training. Entries must be in sequential order.
Y: Response variable vector to model the conditional quantile of, for training. Entries must be in sequential order.
q_level: Probability level of the desired conditional quantiles to predict.
hidden_size: Dimension of the hidden latent state variables in the recurrent network.
num_layers: Number of recurrent layers.
rnn_type: Type of recurrent architecture, can be one of "lstm" (default) or "gru".
p_drop: Probability parameter for dropout before each hidden layer for regularization during training.
learning_rate: Initial learning rate for the optimizer during training of the neural network.
L2_pen: L2 weight penalty parameter for regularization during training.
seq_len: Data sequence length (i.e. number of past observations) used during training to predict each response quantile.
scale_features: Whether to rescale each input covariates to zero mean and unit covariance before applying the network (recommended).
n_epochs: Number of training epochs.
batch_size: Batch size used during training.
X_valid: Covariates in a validation set, or NULL. Entries must be in sequential order. Used for monitoring validation loss during training, enabling learning-rate decay and early stopping.
Y_valid: Response variable in a validation set, or NULL. Entries must be in sequential order. Used for monitoring validation loss during training, enabling learning-rate decay and early stopping.
lr_decay: Learning rate decay factor.
patience_decay: Number of epochs of non-improving validation loss before a learning-rate decay is performed.
min_lr: Minimum learning rate, under which no more decay is performed.
patience_stop: Number of epochs of non-improving validation loss before early stopping is performed.
tol: Tolerance for stopping training, in case of no significant training loss improvements.
fold_separation: Index of fold separation or sequential discontinuity in the data.
warm_start_path: Path of a saved network using torch::torch_save(), to load back for a warm start.
patience_lag: The validation loss is considered to be non-improving if it is larger than on any of the previous patience_lag epochs.
optim_met: DEPRECATED. Optimization algorithm to use during training. "adam" is the default.
seed: Integer random seed for reproducibility in network weight initialization.
verbose: Amount of information printed during training (0:nothing, 1:most important, 2:everything).
device: (optional) A torch::torch_device(). Defaults to default_device().

Value

An QRN object of classes c("QRN_seq", "QRN"), containing the fitted network, as well as all the relevant information for its usage in other functions.