Create MVP summation model

[dfulu]

Pull Request

Description

This is a minimum viable product for this library to train a summation model and log the results to wandb. This inherits the basic structure and a lot of code from PVNet, whilst also trying to minimise duplication (i.e. importing as much as possible rather than copying).

The library includes:

• A base model which includes everything needed to train a model on MAE or multiple quantiles
  • In training, the base class expects to take concurrent PVNet inputs. It puts these through the PVNet model, and then puts the PVNet outputs through the summation model. This could be extended/replaced later if we pre-save PVNet outputs.
• A single minimal model that inherits from the base model
  • This model can use any AbstractLinearNetwork subclass from PVNet to output predictions
  • The inputs are the predictions which come out of PVNet - either point estimates or quantiles. No other inputs are used currently
• A dataloader which loads pre-saved concurrent batches prepared with a script in PVNet
  • The dataloader loads the batches and adds the national output target
• Scripts and configs for training
• Tests

A model from this library is currently being trained and logged to wandb here

Checklist:

• [x] My code follows OCF's coding style guidelines
• [x] I have performed a self-review of my own code
• [x] I have made corresponding changes to the documentation
• [x] I have added tests that prove my fix is effective or that my feature works
• [x] I have checked my code and corrected any misspellings