def test_mgcv_posterior_samples():
formula_gauss = "y ~ x1 + x2 + te(lng, lat, bs='gp', m=list(c(2, 4, 2)), d=2, k=-1)"
formula_binom = "cbind(y, 100 - y) ~ x1 + x2 + te(lng, lat, bs='gp', m=list(c(2, 4, 2)), d=2, k=-1)"
formula_poiss = "y ~ offset(log(n)) + x1 + x2 + te(lng, lat, bs='gp', m=list(c(2, 4, 2)), d=2, k=-1)"
m_gauss = r_methods.mgcv_fit(formula_gauss,
f_data2,
family='gaussian',
weights=None,
method='REML')
m_binom = r_methods.mgcv_fit(formula_binom,
f_data2,
family='binomial',
weights=None,
method='REML')
m_poiss = r_methods.mgcv_fit(formula_poiss,
f_data2,
family='poisson',
weights=None,
method='REML')
link = r_methods.mgcv_posterior_samples(m_binom,
f_data2,
n_samples=100,
response_type='link')
ilink_gauss = r_methods.mgcv_posterior_samples(
m_gauss, f_data2, n_samples=100, response_type='inverse_link')
ilink_binom = r_methods.mgcv_posterior_samples(
m_binom, f_data2, n_samples=100, response_type='inverse_link')
ilink_poiss = r_methods.mgcv_posterior_samples(
m_poiss, f_data2, n_samples=100, response_type='inverse_link')
resp_gauss = r_methods.mgcv_posterior_samples(m_gauss,
f_data2,
n_samples=100,
response_type='response')
assert isinstance(link, np.ndarray)
assert link.shape[0] == 100
assert link.shape[1] == f_data2.shape[0]
assert isinstance(ilink_gauss, np.ndarray)
assert isinstance(ilink_binom, np.ndarray)
assert isinstance(ilink_poiss, np.ndarray)
assert isinstance(resp_gauss, np.ndarray)
with pytest.raises(NotImplementedError):
r_methods.mgcv_posterior_samples(m_binom,
f_data2,
n_samples=100,
response_type='xx')
with pytest.raises(NotImplementedError):
r_methods.mgcv_posterior_samples(m_binom,
f_data2,
n_samples=100,
response_type='response')
def test_mgcv_predict():
formula_binom = "cbind(y, 100 - y) ~ x1 + x2 + te(lng, lat, bs='gp', m=list(c(2, 4, 2)), d=2, k=-1)"
m_binom = r_methods.mgcv_fit(formula_binom,
f_data2,
family='binomial',
weights=None,
method='REML')
response = r_methods.mgcv_predict(m_binom,
f_data2,
response_type='response')
link = r_methods.mgcv_predict(m_binom, f_data2, response_type='link')
lpmatrix = r_methods.mgcv_predict(m_binom,
f_data2,
response_type='lpmatrix')
assert isinstance(response, np.ndarray)
assert isinstance(link, np.ndarray)
assert isinstance(lpmatrix, np.ndarray)
assert response.size == f_data2.shape[0]
assert link.size == f_data2.shape[0]
assert lpmatrix.shape[0] == f_data2.shape[0]
assert lpmatrix.shape[1] == len(rstats.coef(m_binom))
with pytest.raises(NotImplementedError):
r_methods.mgcv_predict(m_binom, f_data2, response_type='xx')
def test_get_link():
formula_binom = "cbind(y, 100 - y) ~ x1 + x2 + te(lng, lat, bs='gp', m=list(c(2, 4, 2)), d=2, k=-1)"
m_binom = r_methods.mgcv_fit(formula_binom,
f_data2,
family='binomial',
weights=None,
method='REML')
link = r_methods.get_link(m_binom)
assert link == 'logit'
def test_summary(capfd):
formula_binom = "cbind(y, 100 - y) ~ x1 + x2 + te(lng, lat, bs='gp', m=list(c(2, 4, 2)), d=2, k=-1)"
m_binom = r_methods.mgcv_fit(formula_binom,
f_data2,
family='binomial',
weights=None,
method='REML')
r_methods.summary(m_binom)
out, err = capfd.readouterr()
assert out is not None
def test_get_names():
rdframe = r_methods.pdframe2rdframe(f_data)
obj_names1 = r_methods.get_names(rdframe)
assert isinstance(obj_names1, list)
formula_binom = "cbind(y, 100 - y) ~ x1 + x2 + te(lng, lat, bs='gp', m=list(c(2, 4, 2)), d=2, k=-1)"
m_binom = r_methods.mgcv_fit(formula_binom,
f_data2,
family='binomial',
weights=None,
method='REML')
obj_names2 = r_methods.get_names(m_binom)
assert isinstance(obj_names2, list)
def test_mgcv_fit():
formula_gauss = "y ~ x1 + x2 + te(lng, lat, bs='gp', m=list(c(2, 4, 2)), d=2, k=-1)"
formula_binom = "cbind(y, 100 - y) ~ x1 + x2 + te(lng, lat, bs='gp', m=list(c(2, 4, 2)), d=2, k=-1)"
formula_poiss = "y ~ offset(log(n)) + x1 + x2 + te(lng, lat, bs='gp', m=list(c(2, 4, 2)), d=2, k=-1)"
weights = np.random.uniform(.1, 1, f_data2.shape[0])
m_binom = r_methods.mgcv_fit(formula_binom,
f_data2,
family='binomial',
weights=None,
method='REML')
m_poiss = r_methods.mgcv_fit(formula_poiss,
f_data2,
family='poisson',
weights=None,
method='REML')
m_gauss = r_methods.mgcv_fit(formula_gauss,
f_data2,
family='gaussian',
weights=weights,
method='REML')
m_gauss_bam = r_methods.mgcv_fit(formula_gauss,
f_data2,
family='gaussian',
weights=None,
method='fREML',
bam=True)
#m_gaus_wam = r_methods.mgcv_fit(formula_gauss, f_data2, family='gaussian', weights=weights.tolist(), method='fREML',
# bam=True)
assert isinstance(m_binom, robjects.vectors.ListVector)
assert isinstance(m_poiss, robjects.vectors.ListVector)
assert isinstance(m_gauss, robjects.vectors.ListVector)
assert isinstance(m_gauss_bam, robjects.vectors.ListVector)
#assert isinstance(m_gauss_wam, robjects.vectors.ListVector)
with pytest.raises(NotImplementedError):
r_methods.mgcv_fit(formula_binom,
f_data2,
family='xx',
weights=None,
method='REML')
with pytest.raises(NotImplementedError):
r_methods.mgcv_fit(formula_binom,
f_data2,
family='xx',
weights=weights[:40],
method='REML')
with pytest.raises(NotImplementedError):
r_methods.mgcv_fit(formula_binom,
f_data2,
family='xx',
weights=weights[:, None],
method='REML')
with pytest.raises(NotImplementedError):
r_methods.mgcv_fit(formula_binom,
f_data2,
family='xx',
weights=list(weights),
method='REML')