In [1]: import statsmodels.datasets as datasets
In [2]: import sklearn.metrics as metrics
In [3]: from numpy import log
In [4]: from pyearth import Earth as earth
In [5]: boston = datasets.get_rdataset("Boston", "MASS").data
In [6]: x = boston.ix[:, 0:boston.shape[1] - 1]
In [7]: xlabel = list(x.columns)
In [8]: y = boston.ix[:, boston.shape[1] - 1]
In [9]: model = earth(enable_pruning = True, penalty = 3, minspan_alpha = 0.05, endspan_alpha = 0.05)
In [10]: model.fit(x, log(y), xlabels = xlabel)
Out[10]:
Earth(allow_linear=None, check_every=None, enable_pruning=True, endspan=None,
endspan_alpha=0.05, max_degree=None, max_terms=None,
min_search_points=None, minspan=None, minspan_alpha=0.05, penalty=3,
smooth=None, thresh=None)
In [11]: print model.summary()
Earth Model
--------------------------------------
Basis Function Pruned Coefficient
--------------------------------------
(Intercept) No 2.93569
h(lstat-5.9) No -0.0249319
h(5.9-lstat) No 0.067697
h(rm-6.402) No 0.230063
h(6.402-rm) Yes None
crim Yes None
h(dis-1.5004) No -0.0369272
h(1.5004-dis) No 1.70517
h(ptratio-18.6) No -0.0393493
h(18.6-ptratio) No 0.0278253
nox No -0.767146
h(indus-25.65) No -0.147471
h(25.65-indus) Yes None
h(black-395.24) Yes None
h(395.24-black) Yes None
h(crim-24.8017) Yes None
h(24.8017-crim) No 0.0292657
rad No 0.00807783
h(rm-7.853) Yes None
h(7.853-rm) Yes None
chas Yes None
--------------------------------------
MSE: 0.0265, GCV: 0.0320, RSQ: 0.8409, GRSQ: 0.8090
In [12]: metrics.r2_score(log(y), model.predict(x))
Out[12]: 0.840861083407211
Yet Another Blog in Statistical Computing 