Explaining class assignment objectives for data frame analytics jobs

[valeriy42]

This PR adds a notebook that explains the properties of different class assignment objectives available for the machine learning data frame analytics classification jobs and gives pointers on when to choose which objective.

It intends to be linked in the "Concepts" section of the documentation and provide users with more insights than it is possible in the scope of "Concepts".

The rendered version of the jupyter notebook can be found here.

Closes #356.

[valeriy42]

@szabosteve and @lcawl it would be great if one of you could take a look if the explanations make sense to you. 🙏

[tveasey]

All in all, examples and discussion looks spot on to me, I only raised a couple of typos. No further review needed from me.

[szabosteve]

Thanks, @valeriy42 for this example and for the explanation. I left some minor comments, mostly nitpicking. Please take or leave them as you want.

[szabosteve]

Suggested change

	"* The default objective `maximize_minimum_recall` should work well for most cases.\n",
	"* The default objective `maximize_minimum_recall` works well for most cases.\n",

[szabosteve]

Suggested change

	"* You should explicitly use the `maximize_minimum_recall` objective if you have an unbalanced dataset and you are interested in high recall on the rare class.\n",
	"* Use the `maximize_minimum_recall` objective if you have an unbalanced dataset and you are interested in high recall on the rare class.\n",

[szabosteve]

Suggested change

	"* You should use the `maximize_accuracy` objective if you have a balanced dataset and you are interested in high accuracy and precision.\n",
	"* Use the `maximize_accuracy` objective if you have a balanced dataset and you are interested in high accuracy and precision.\n",

[szabosteve]

Suggested change

	"There is a trade-off between maximizing precision and recall. For example, labeling all people as \"infected\" will get us 100% but recall for infected people, since we identified all infected people in the sample, but only a very small precision since we misclassified many healthy people as \"infected\" as well.\n",
	"There is a trade-off between maximizing precision and recall. For example, labeling all people as \"infected\" will get us 100% recall for infected people, since we identified all infected people in the sample, but only a very small precision since we misclassified many healthy people as \"infected\" as well.\n",

[szabosteve]

This section is handy, I really like it!

[szabosteve]

Nit. We prefer to use data set in the docs, but dataset is also correct.

Suggested change

	"We begin by looking at a balanced classification dataset where the number of data points for class 0 and class 1 are about the same. "
	"We begin by looking at a balanced classification data set where the number of data points for class 0 and class 1 are about the same. "

[szabosteve]

It is not rendered correctly in the output.