Revisit evolvable parameters for existing classifier operators

[rhiever]

Here are some parameters that we should consider adding to the existing classifier operators.

SVC is currently removed from TPOT, and we may remove it entirely because it requires too much parameter tuning to perform well.

DecisionTreeClassifier

http://scikit-learn.org/stable/modules/generated/sklearn.tree.DecisionTreeClassifier.html

Existing evolvable parameters: max_features, max_depth

New evolvable parameters: criterion

RandomForestClassifier

http://scikit-learn.org/stable/modules/generated/sklearn.ensemble.RandomForestClassifier.html

Existing evolvable parameters: max_features

New evolvable parameters: criterion

LogisticRegression

http://scikit-learn.org/stable/modules/generated/sklearn.linear_model.LogisticRegression.html

Existing evolvable parameters: C

New evolvable parameters: fit_intercept, dual, penalty

SVC

http://scikit-learn.org/stable/modules/generated/sklearn.svm.SVC.html

Existing evolvable parameters: C

New evolvable parameters: kernel, gamma, coef0

KNeighborsClassifier

http://scikit-learn.org/stable/modules/generated/sklearn.neighbors.KNeighborsClassifier.html

Existing evolvable parameters: n_neighbors

New evolvable parameters: none

[rhiever]

@teaearlgraycold, forget this issue for now. I will have an updated version of it in a separate issue soon.

[danthedaniel]

Alright. I had not started work on it yet so that's fine.