Colin Catlin

Current logo doesn't work well with dark mode.

As described. Although whether to do this only with model_forecast or with full AutoTS class (fractional validation?) is an open question

mosaic window, max models for horizontal, clusters/subsample count

This will effect "seasonal" and "similarity" methods

If incoming index is another datatype format or a coerce-able string, attempt to convert to DatetimeIndex

Simplify users having to choose a model list. Set as default.

Speed Up Uncertainty Predictions

12

Addresses #2030 with the following approach: maintains functionality of `predict_uncertainty` and `sample_posterior_predictive` while making as obsolete `sample_model` and `sample_predictive_trend`. However, this allows the full original approach to uncertainty when `mcmc_sampling...

I might suggest enabling [GitHub Discussions](https://docs.github.com/en/discussions) page with the hope of putting all of the issues of the sort "Prophet isn't working for my dataset, please help" over there and...

Or else the typing import needs to be removed (it isn't strictly necessary for any functionality). I can open a PR for this, it's pretty simple, but I think someone...

ENH Add Additional Losses to MLPRegressor

12

This issue adds additional losses of MAE and Logcosh to the MLPRegressor. Yes, yes, I know that Torch/Tensorflow/etc are the place where 'real' neural networks happen, but as I work...