def setup_class(cls):
# simulate data
np.random.seed(987865)
nobs, k_vars = 500, 10
k_nonzero = 4
x = ((np.random.rand(nobs, k_vars) +
0.5* (np.random.rand(nobs, 1) - 0.5)) * 2 - 1)
x *= 1.2
x[:, 0] = 1
beta = np.zeros(k_vars)
beta[:k_nonzero] = 1. / np.arange(1, k_nonzero + 1)
linpred = x.dot(beta)
y = cls._generate_endog(linpred)
cls.k_nonzero = k_nonzero
cls.x = x
cls.y = y
# defaults to be overwritten by subclasses
cls.rtol = 1e-4
cls.atol = 1e-6
cls.exog_index = slice(None, None, None)
cls.k_params = k_vars
cls.skip_hessian = False # can be overwritten in _initialize
cls.penalty = smpen.SCADSmoothed(0.1, c0=0.0001) # default for tests
cls._initialize()
def __init__(self, *args, **kwds):
super(PenalizedMixin, self).__init__(*args, **kwds)
penal = kwds.pop('penal', None)
# I keep the following instead of adding default in pop for future changes
if penal is None:
# TODO: switch to unpenalized by default
self.penal = smpen.SCADSmoothed(0.1, c0=0.0001)
else:
self.penal = penal
# TODO: define pen_weight as average pen_weight? i.e. per observation
# I would have prefered len(self.endog) * kwds.get('pen_weight', 1)
# or use pen_weight_factor in signature
self.pen_weight = kwds.get('pen_weight', len(self.endog))
self._init_keys.extend(['penal', 'pen_weight'])
def setup_class(cls):
x0 = np.linspace(-0.2, 0.2, 11)
cls.params = np.column_stack((x0, x0))
cls.pen = smpen.SCADSmoothed(tau=0.05, c0=0.05)
