Function reference • aion

Layers

layer_glu(): Gated Linear Unit

layer_grn(): Gated Residual Network block

layer_lmu(): Legendre Memory Unit layer

layer_multi_dense(): Multiple dense layers in one layer

layer_multi_embedding(): Multiple embeddings in one layer

layer_tcn(): Residual block for the WaveNet TCN

layer_vsn(): Variable Selection Network block

layer_scaled_dot_attention(): Scaled dot product attention layer

layer_interpretable_mh_attention(): Interpretable multi-head attention layer

layer_temporal_fusion_decoder(): Temporal Fusion Decoder layer

Losses

loss_quantile() loss_pinball(): Quantile (Pinball) loss function

loss_tweedie(): Tweedie Loss (negative log likelihood)

loss_negative_log_likelihood(): General negative log likelihood loss function

Metrics

metric_q_risk(): q-Risk metric

Models

model_tft(): Temporal Fusion Transformer model

Data utils

make_arrays(): Prepare input/taget arrays for time series models

ts_generator(): Create a generator for time series data

neaten(): Construct a data.frame out of the output forecast tensor

dict_size(): Get cardinalities of the categorical variables

rand_array(): Generate an array of random values with the given dimensions

find_gaps(): Check if time series in the panel dataset contain time gaps.