Package index • EQRN

Fitting EQRN Tail Neural Networks

Functions to fit a tail GPD EQRN network to intermediate quantile exceedences

EQRN_fit_restart(): Wrapper for fitting EQRN with restart for stability

EQRN_fit(): EQRN fit function for independent data

EQRN_fit_seq(): EQRN fit function for sequential and time series data

Predicting using fitted EQRN networks

EQRN_predict(): Predict function for an EQRN_iid fitted object

EQRN_predict_seq(): Predict function for an EQRN_seq fitted object

EQRN_predict_params(): GPD parameters prediction function for an EQRN_iid fitted object

EQRN_predict_params_seq(): GPD parameters prediction function for an EQRN_seq fitted object

EQRN_excess_probability(): Tail excess probability prediction using an EQRN_iid object

EQRN_excess_probability_seq(): Tail excess probability prediction using an EQRN_seq object

compute_EQRN_GPDLoss(): Generalized Pareto likelihood loss of a EQRN_iid predictor

compute_EQRN_seq_GPDLoss(): Generalized Pareto likelihood loss of a EQRN_seq predictor

Saving and Loading

EQRN_save(): Save an EQRN object on disc

EQRN_load(): Load an EQRN object from disc

Neural Networks Helpers

install_backend(): Install Torch Backend

default_device(): Default torch device

loss_GPD_tensor(): GPD tensor loss function for training a EQRN network

quantile_loss_tensor(): Tensor quantile loss function for training a QRN network

get_excesses(): Computes rescaled excesses over the conditional quantiles

process_features(): Feature processor for EQRN

perform_scaling(): Performs feature scaling without overfitting

mts_dataset(): Dataset creator for sequential data

FC_GPD_net(): MLP module for GPD parameter prediction

FC_GPD_SNN(): Self-normalized fully-connected network module for GPD parameter prediction

Separated_GPD_SNN(): Self-normalized separated network module for GPD parameter prediction

Recurrent_GPD_net(): Recurrent network module for GPD parameter prediction

QRNN_RNN_net(): Recurrent quantile regression neural network module

Extreme Value Analysis Helpers

GPD_excess_probability(): Tail excess probability prediction based on conditional GPD parameters

fit_GPD_unconditional(): Maximum likelihood estimates for the GPD distribution using peaks over threshold

predict_unconditional_quantiles(): Predict unconditional extreme quantiles using peaks over threshold

predict_GPD_semiconditional(): Predict semi-conditional extreme quantiles using peaks over threshold

loss_GPD(): Generalized Pareto likelihood loss

unconditional_train_valid_GPD_loss(): Unconditional GPD MLEs and their train-validation likelihoods

semiconditional_train_valid_GPD_loss(): Semi-conditional GPD MLEs and their train-validation likelihoods

GPD_quantiles(): Compute extreme quantile from GPD parameters

Intermediate Models

QRN_seq_fit(): Recurrent QRN fitting function

QRN_fit_multiple(): Wrapper for fitting a recurrent QRN with restart for stability

QRN_seq_predict(): Predict function for a QRN_seq fitted object

QRN_seq_predict_foldwise(): Foldwise fit-predict function using a recurrent QRN

QRN_seq_predict_foldwise_sep(): Sigle-fold foldwise fit-predict function using a recurrent QRN

Accuracy metrics

mean_squared_error(): Mean squared error

mean_absolute_error(): Mean absolute error

square_loss(): Square loss

quantile_loss(): Quantile loss

prediction_bias(): Prediction bias

prediction_residual_variance(): Prediction residual variance

R_squared(): R squared

proportion_below(): Proportion of observations below conditional quantile vector

quantile_prediction_error(): Quantile prediction calibration error

quantile_exceedance_proba_error(): Quantile exceedance probability prediction calibration error

multilevel_MSE(): Multilevel quantile MSEs

multilevel_MAE(): Multilevel quantile MAEs

multilevel_q_loss(): Multilevel quantile losses

multilevel_pred_bias(): Multilevel prediction bias

multilevel_resid_var(): Multilevel residual variance

multilevel_R_squared(): Multilevel R squared

multilevel_prop_below(): Multilevel 'proportion_below'

multilevel_q_pred_error(): Multilevel 'quantile_prediction_error'

multilevel_exceedance_proba_error(): Multilevel 'quantile_exceedance_proba_error'

Other Helpers

check_directory(): Check directory existence

safe_save_rds(): Safe RDS save

last_elem(): Last element of a vector

roundm(): Mathematical number rounding

vec2mat(): Convert a vector to a matrix

make_folds(): Create cross-validation folds

lagged_features(): Covariate lagged replication for temporal dependence

vector_insert(): Insert value in vector

get_doFuture_operator(): Get doFuture operator

set_doFuture_strategy(): Set a doFuture execution strategy

end_doFuture_strategy(): End the currently set doFuture strategy

Prediction methods

S3 class method support for classes EQRN_iid, EQRN_seq and QRN_seq, and methods predict and excess_probability, as a facultative alternative to their respective functions above.

predict(<EQRN_iid>): Predict method for an EQRN_iid fitted object

predict(<EQRN_seq>): Predict method for an EQRN_seq fitted object

predict(<QRN_seq>): Predict method for a QRN_seq fitted object

excess_probability(): Excess Probability Predictions

excess_probability(<EQRN_iid>): Tail excess probability prediction method using an EQRN_iid object

excess_probability(<EQRN_seq>): Tail excess probability prediction method using an EQRN_iid object