To close #6575 and
#6684.

The notebook is originally from
#6551.

Copy some discussion:

---

fkiraly:

Great! FYI @benHeid, have you seen and reviewed this?

I mainly have comments about the notebook.

* could you separate this into another PR? The notebook is great, but
there is some iteration needed regarding location and integration with
the other notebooks. I think we can merge the forecasters already and
don't need to delay.

Some minor comments on the notebook content:
* there are a lot of printouts that confuse the reader. These should be
silenced so you can focus on the didactic content.
* the markdown text is nice, I would just format it so the lines are not
too long, and I would also use shorter telegram style, like on ppt
slides

Regarding location, I would actually add the content to the notebook
01c, which has some content already. There is some minor confusion in
the notebook about terminology, the notebook also uses the term "global"
but in a different way.
* in the M5 paper, what the 01c notebook does is called "cross-learning"
* the "global" in current 01d is more of a pre-training on other
instances

In any case, we need to disambiguate terminology and perhaps adopt
clearer distinctions here. @benHeid, what do you suggest on how we
handle the terminology clash between 01c and 01d? And, should this go in
the same notebook, so the "multiple instances" cases can be explained
easily?

_Originally posted by @fkiraly in
#6551 (review)
            
---

shlok191:

@fkiraly, @Xinyu-Wu-0000. Yes, I have some minor input! In the
`01d_forecasting_global_forecast.ipynb` notebook and the
`01c_forecasting_hierarchical_global.ipynb` notebooks, "global learning"
is referred to as a term. I do agree that instead of using "global
learning", we can instead use "pretrained" and "cross-learning" as
replacements.

Here is how I'd differentiate them:
- To the best of my knowledge, pre-trained models do not require the
training dataset time series to be correlated to each other

- Referencing this [paper on the M5
competition](https://www.sciencedirect.com/science/article/pii/S0169207021001874),
I believe "cross-learning" is the term utilized for training models on
time series which have a strong correlation. I think M5 was referenced
in 01c, so it might be good to use "cross-learning" there instead of
"global learning"!

_Originally posted by @shlok191 in
#6551 (comment)