`matplotlib` Unit Tests; Test Coverage

[rosenbrockc]

I still haven't got any unit tests for the decorated matplotlib. I ran into "not installed as a framework" errors; travis is giving problems with the packages at the moment anyway, so I am just leaving it for a while. This accounts for some of the low test coverage.

The other low coverage is from the descriptors machinery (see #16).

The analysis methods for sklearn only have a classifier example at the moment. They are also extremely basic and don't take any of the best practices into account. We need to update this to be useful.

We can have a discussion about analysis methods for other common class instances and data types.

If we hit these three points, the test coverage would be up to 100% easily.

[andrewhuynguyen]

I can help with both the descriptors machinery and sklearn. I have some other methods (regression, classification, etc.) for sklearn that I could add to acorn.

[rosenbrockc]

Awesome. If you look at acorn.analyze.sklearn module, you will see a routine fit and another one called predict. The first argument is the FQDN of the method (for example sklearn.svm.base.fit), the rest are described in the API docs. If you add additional methods to this module, you can tie them in under classify_fit, regress_fit, classify_predict and regress_predict. If you look through the whole module, it should be pretty obvious what's going on.

If you want to add a more direct handler for one of the methods, then add it's FQDN under the [analysis.methods] section of sklearn.cfg. There are some examples in that file already that point to the acorn.analyze.sklearn module.