The ``NOPScaler`` assigns a scale equals to 1 to each input item, i.e., no scaling is applied upon calling the ``NOPScaler``.

A multi-GPU-friendly python command line logger.

The ``MeanScaler`` computes a per-item scale according to the average absolute value over time of each item. The average is computed only amo

The ``MeanStdScaler`` computes a per-item scale according to the mean and standard deviation over time of each item. The mean and standard

The traditional encoder for TACTiS, based on the Transformer architecture. The encoder receives an input which contains for each series and

MADE building block layer

A class implementing the positional encoding for Transformers described in Vaswani et al. (2017). Somewhat generalized to allow unaligned or

A unified sampler for generating stochastic process trajectories including: - Brownian Motion - Brownian Bridge - Autoregressive

The encoder for TACTiS, based on the Temporal Transformer architecture. This encoder alternate between doing self-attention between different

A non-parametric copula based on attention between the various variables.

RealNVP BatchNorm layer

Container for layers of a normalizing flow

The MultivariateEvaluator class evaluates forecasts for multivariate or multi-dimensional observations. Evaluations of individual dimens

The compatibility layer between TACTiS and GluonTS / PyTorchTS.

A decoder which forecast using a distribution built from a copula and marginal distributions.

Compute the marginals using a Deep Sigmoid Flow conditioned using a MLP. The conditioning MLP uses the embedding from the encoder as its inpu

Modified RealNVP Coupling Layers per the MAF paper

Randomly pick a single valid window in the given time series. This fix the bias in ExpectedNumInstanceSampler which leads to varying sampling

The top-level module for TACTiS. The role of this module is to handle everything outside of the encoder and decoder. This consists mainl

A shell on top of the TACTiS module, to be used during training only. loss ?

Evaluator class, to compute accuracy metrics by comparing observations to forecasts. Parameters ---------- quantiles lis

Trivial normalization helper. Do nothing to its data.

Normalization helper for the standardization. The data for each batch and each series will be normalized by: - substracting the historic

An `Estimator` type with utilities for creating PyTorch-Lightning-based models. To extend this class, one needs to implement three metho