Adding very short section on conventions.

[amueller]

Just saw #1679 again and thought we should add some "conventions" documentation.
I think adding a short paragraph about sklearn rules and expectations to the "quick start guide" would be nice.

It should contain "unless otherwise specified, input will be cast to float64", regression targets will be converted to float64, classification targets can be arbitrary and the same type will be produced again.
Also maybe in the same place say that calling fit will forget any previous models, and maybe that parameters can be set using estimator.parameter = stuff?

I have no other ideas what should be there.

[cangermueller]

@amueller, is this the "quick start guide"? I can create a convention section about the fact that

• input will be cast to float64, and
• fit() overwrites the state of previous models

Can you elaborate on what you mean with estimator.paramter = stuff?

[amueller]

The link you gave is very old (version 0.10). I meant this: http://scikit-learn.org/dev/tutorial/basic/tutorial.html
Which you can find on the documentation page under "Quick start". It is indeed the newer version of the page you linked to.

I meant that you can set parameters on an object before fitting without reconstructing the object.

[amueller]

closed in 0fe613e