Config file algorithm specification

[ceholden]

Make way for more timeseries algorithms within module by changing configuration file to be able to point to many different algorithms:

1. Add new submodule, algorithms
2. Rename yatsm to ccdc and place into algorithms submodule. YATSM class to CCDCesque
3. Change configuration file by adding algorithm key under YATSM section. The algorithm specified by algorithm key will be searched for as the section title from which to extract algorithm parameterization information.
4. Add new YATSM section for options generic to all timeseries algorithms, like reverse or robust.
5. Remove robust results and omission and commission tests from current YATSM (future, CCDCesque) and place into yatsm.algorithms.yatsm. These will be parameterized in YATSM metadata section.
6. Add section for regression/predictive model method configuration (see Model prediction / regression selection #26)

Propose change example:

[metadata]
version = 0.5

[YATSM]
algorithm = CCDCesque
regression = LassoCV
design_matrix = 1 + x + harm(x, 1)
reverse = False
robust = False
commission_alpha = 
...

[CCDCesque]
consecutive = 5

[LassoCV]
pickle = somefile.pkl
...

It is very difficult to imagine specifying all arguments to a sklearn classifier or regression estimator via a config file. Things like n_alpha could play well, but how would we specify alphas = np.logspace(0.001, 30, 50)? This proposed format sidesteps these concerns by requiring that regression options provide a pickled file from sklearn.external.joblib that already contains the parameterization desired. If the pickle item is not provided, but the section is labeled, default to a pickle of an existing regression object packaged with yatsm.

Target v0.5.0 as milestone to coincide with another major rehaul (#28).

[ceholden]

Also now using YAML! See #30