Cristian Challu

Transformer models need calendar variables and they do not admit additional exogenous variables. For them to work in different settings than in the benchmark datasets they should: 1. Not always...

**Is your feature request related to a problem? Please describe.** A clear and concise description of what the problem is. Ex. I'm always frustrated when [...] **Describe the solution you'd...

[Models] Evaluate `torch.compile` on `nf` models

7

### Description The `torch.compile` method has been introduced in PyTorch 2.0 and is aimed at optimizing and accelerating the execution of PyTorch models. The NeuralForecast library, which leverages PyTorch for...

[Models] Add feature importance estimation

2

### Description Add estimation of the importance of exogenous covariates, integrating explainability algorithms such as SHAP or LIME. ### Use case _No response_

### What happened + What you expected to happen Wrong initialization of embeddings, use historic size. ### Versions / Dependencies all ### Reproduction script None ### Issue Severity Medium: It...

[Transformers] Improve exogenous covariates capabilities

1

### Description Current transformer models (`PatchTST`, `Informer`, `Autoformer`, `FEDformer`) only allow for calendar covariates. Adding the capability to include continuous temporal or static covariates would improve their performance on many...

### Description Add validation set dataframe to `core` methods. The loader will take this dataset to sample series for validation. ### Use case _No response_

### Description Split models in `test-model-performance` into several separate tests to speed up evaluation. ### Use case _No response_

### Description As requested in #579, add the possibility to split train/val/test sets by date rather than the number of timestamps or the number of windows. ### Use case _No...