Ridge multiple lags

[jannesvaningen]

Some last minute fixes for reading presentation in:

• readme for installation
• s2spy added to dependencies
• path of model outputs
• timeseries plots for lstm

• Added a lags for loop in the ridge ipynb to accomodate fitting a ridge model for multiple lags add once.
  I also added a line plot to show the y_train and y_test outcomes of the model.

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[geek-yang]

Line #8.        f"The MSE of climatology is {mean_squared_error(ground_truth, np.repeat(target_clim, ground_truth.anchor_year.size)):.3f}"

Thanks for correcting it. I forgot to update this with climatology. Btw, for a fair evaluation, I think when calculating the climatology, we shall only include the training data, instead of the whole dataset (if you are looking for something else, just ignore my comment).

Reply via ReviewNB

[geek-yang]

Thanks a lot for adding the multi lag regression to our recipes for a fair comparison. I planned to do it but didn't manage to find time. 😉

Just a small comment, it seems the deep learning ones are not very robust. Maybe we can add random seeds, to make the results reproducible, for instance:

torch.manual_seed(1)
# for python related randomness
random.seed(1)
# for numpy related randomness
np.random.seed(1)

(more information https://pytorch.org/docs/stable/notes/randomness.html)

Nice work @jannesvaningen !

[semvijverberg]

I check all notebooks, fixed some merge issue related to later updates (fixed seed + downloading data).

+Now computing mean_squared_error for every notebook in the same manner.