A simple way to access a small classic dataset from datatable

[oleksiyskononenko]

What if... we bring several small classic datasets to datatable, so that we could use them in our tutorials/tests by simply doing something like from datatable import DT_iris, DT_credit_card.

[st-pasha]

I would prefer not to pollute the global namespace with too many names, so how about we have a function that loads the named dataset, like

dt.example('iris')
dt.example('creditcard')

[oleksiyskononenko]

Yes, the access could be whatever we think is relevant... I've been also thinking about simplifying access to the classic datasets in general through smth like dt.load('airlines'), but it seems that there are no standard names and URLs where these datasets could be found.

[st-pasha]

We need a function name that makes it clear that it's for sample datasets only, not for loading generic data. So dt.load() is not good. On the other hand, i don't quite like dt.example() either...

[oleksiyskononenko]

dt.load() was a different thing, it was another idea of having a simple way to load a classic dataset from the web... but when I checked, there are no standard names for these datasets, so this idea makes no sense.

[st-pasha]

There is a public repository of various datasets: https://archive.ics.uci.edu/ml/datasets.php
There are lots of examples there, so we might as well choose something that has not yet bored everyone to death :)

[oleksiyskononenko]

Yeah, we definitely have a choice as for the small datasets for tutorials :) I would even use one in the FTRL tests... but if I want to have a shortcut to load airlines or any other dataset, like through a possible dt.load() I have mentioned above, it could become kind of messy. It is just simpler to load the data by URL.

[st-pasha]

For example this one looks good: https://archive.ics.uci.edu/ml/machine-learning-databases/flags/flag.data