def test_knot_candidates():
np.random.seed(10)
m = 1000
x = np.random.normal(size=m)
p = np.random.normal(size=m)
p[np.random.binomial(p=.1, n=1,size=m)==1] = 0.
x[np.random.binomial(p=.1, n=1,size=m)==1] = 0.
predictor = PredictorDependentData.alloc(x)
candidates, candidates_idx = predictor.knot_candidates(p, 5, 10, 0, 0, set())
assert_array_equal(candidates, x[candidates_idx])
assert_equal(len(candidates), len(set(candidates)))
def test_knot_candidates():
np.random.seed(10)
m = 1000
x = np.random.normal(size=m)
p = np.random.normal(size=m)
p[np.random.binomial(p=.1, n=1, size=m) == 1] = 0.
x[np.random.binomial(p=.1, n=1, size=m) == 1] = 0.
predictor = PredictorDependentData.alloc(x)
candidates, candidates_idx = predictor.knot_candidates(
p, 5, 10, 0, 0, set())
assert_array_equal(candidates, x[candidates_idx])
assert_equal(len(candidates), len(set(candidates)))
def form_inputs(x, B, p, knot, candidates, y, w):
# Formulate the inputs for the fast version
m, q = B.shape
max_terms = q + 2
workings = []
n_outcomes = w.shape[1]
for _ in range(n_outcomes):
working = KnotSearchWorkingData.alloc(max_terms)
workings.append(working)
outcome = MultipleOutcomeDependentData.alloc(y, w, m, n_outcomes, max_terms, 1e-16)
for j in range(B.shape[1]):
outcome.update_from_array(B[:, j])
predictor = PredictorDependentData.alloc(x)
constant = KnotSearchReadOnlyData(predictor, outcome)
return KnotSearchData(constant, workings, q)
def form_inputs(x, B, p, knot, candidates, y, w):
# Formulate the inputs for the fast version
m, q = B.shape
max_terms = q + 2
workings = []
n_outcomes = w.shape[1]
for _ in range(n_outcomes):
working = KnotSearchWorkingData.alloc(max_terms)
workings.append(working)
outcome = MultipleOutcomeDependentData.alloc(
y, w, m, n_outcomes, max_terms, 1e-16)
for j in range(B.shape[1]):
outcome.update_from_array(B[:, j])
predictor = PredictorDependentData.alloc(x)
constant = KnotSearchReadOnlyData(predictor, outcome)
return KnotSearchData(constant, workings, q)
