Jack Kelly

## Detailed Description Very interesting new paper from Google: * [Blog: Improving Vision Transformer Efficiency and Accuracy by Learning to Tokenize](https://ai.googleblog.com/2021/12/improving-vision-transformer-efficiency.html) * [Paper: TokenLearner: What Can 8 Learned Tokens Do...

**Describe the bug** ``` ______________________________________________________________________________________________________________ test_train _________________________________________________________________________________________________________________ configuration_conv3d = Configuration(general=General(name='example', description='example configuration'), input_data=InputData(pv=PV(start_d...atches=250, n_validation_batches=0, n_test_batches=10, upload_every_n_batches=16, local_temp_path='~/temp/'), git=None) def test_train(configuration_conv3d): config_file = "tests/configs/model/conv3d_sat_nwp.yaml" config = load_config(config_file) dataset_configuration = configuration_conv3d dataset_configuration.input_data.nwp.nwp_image_size_pixels...

Try without historical GSP PV

3

We should see how PV forecasting performance changes if we don't have historical GSP PV. The GSP PV data we'll have at inference time will be less accurate than the...

## Detailed Description Combing multiple ML models often wins Kaggle competitions. The ensemble could include: * Model which takes single step of satellite imagery from optical flow and predicts PV...

During validation, pass all the PV IDs through the embedding, and then through tSNE, and plot. Hopefully we'll see outliers, and these might be "bad" PV systems.

Try 2d, 3d CNN before attention

6

related: #35 3D cnn: sees two timesteps at once

Stop giving the ML model satellite images of the _future_!

2

To test things, we showed the ML model satellite images of the _future_ (i.e we showed the ML model satellite images _after_ `t0`). We should probably stop doing that now...

Format docstrings so they pass `pydocstyle`'s criteria

8

## Detailed Description We plan to use [`pydocstyle`](http://www.pydocstyle.org/en/stable/) to automatically check that our docstrings conform to a standard format. It would be amazing to have help tweaking our existing docstrings...

National Grid ESO are interested in having an improved forecast for National Electricity Demand. And, because all British solar is on the distribution grid, solar generation looks like a reduction...

Log more metrics

2

- [x] Log timeseries of validation data (predicted PV over several hours, and actual PV) - [ ] Separately, create video of the clouds for that validation sequence? Or maybe...