Lorenzo Stella

Move to pyproject.toml

2

*Description of changes:* the objective is to move all of the project config and metadata from setup.py and setup.cfg to pyproject.toml By submitting this pull request, I confirm that you...

Consolidate R methods

4

There are currently [three versions](https://github.com/awslabs/gluonts/blob/dev/src/gluonts/ext/r_forecast/R/univariate_forecast_methods.R#L44-L125) of ARIMA methods: - vanilla - with seasonal features - with custom dynamic features and optionally sesasonal ones The latter requires [special handling](https://github.com/awslabs/gluonts/blob/78353021e7644ae53e26dcea1adce2ddef5756aa/src/gluonts/ext/r_forecast/_univariate_predictor.py#L141) in the...

Add torch version of MQ-CNN and MQ-RNN

1

*Issue #, if available:* Addressing #2059 *Description of changes:* Adding PyTorch version of MQ-CNN and MQ-RNN, with lightning module around it for training purposes. Work in progress, notable features missing:...

PyTorch implementation of MQ-CNN and MQ-RNN

1

References: - https://arxiv.org/pdf/1711.11053.pdf - [MXNet implementation](https://github.com/awslabs/gluon-ts/tree/4126386da1c71a371a77fe824e5092645dc2d2db/src/gluonts/model/seq2seq) (in my opinion this is a little too dispersive) Milestones for this one: - [ ] Implementing core modules (with option for different encoders,...

Extend `Predictor` interface to support variable prediction length

4

## Description Currently, predictors are configured with a fixed `prediction_length` at construction time, and they will only yield forecasts of that length. This is somewhat limiting, and we could have...

Add more flexible sampler types through `Range`

1

*Description of changes:* This proposes a new type of samplers, that select (not necessarily) random instances to construct training or validation batches. The main addition to the existing samplers, is...

`lagged_sequence_values` helper does not work well with multivariate data

7

one issue i found with the `lagged_sequence_values` helper is that with multivariate inputs this helper doesn't return the stacked tensors... e.g. ```py # prior_input.shape torch.Size([1, 167, 137]) # input.shape torch.Size([1,...

Remove SuiteSparse from dependencies

3

Suite Sparse is been merged into SparseArrays. - Removing dependency on SuiteSparse in favor of SparseArrays - Bump minimum Julia requirement to 1.9 - Adjust GH workflows accordingly - Update...

*Description of changes:* Add torch implementation of MQ-CNN. Replaces #2342 TODOs: - [ ] fix linting and type checking issues - [ ] fix and enable commented test cases By...

Factoring loss out of distribution objects

6

Problem: It is not clear how one can select a different loss to train a model, other than by customizing the network's code. (Some proposal/discussion on this problem is also...