def test_transpose_compatible(self):
"""Test that the results of `transpose` are compatible for sparse and dense arrays."""
arr = tocoo(np.arange(27).reshape(3, 3, 3))
np.testing.assert_array_equal(todense(transpose(arr, (1, 2, 0))),
transpose(todense(arr), (1, 2, 0)))
for _ in range(5):
ndims = np.random.randint(2, 6)
dims = tuple(np.random.randint(5, 20, size=ndims))
axes = np.random.permutation(range(ndims))
arr = sp.random(dims, density=0.1)
out1 = todense(transpose(arr, axes))
out2 = transpose(todense(arr), axes)
np.testing.assert_allclose(out1, out2, verbose=True)
def predict_interval(self, X, alpha=.05):
"""
Get a confidence interval for the prediction at `X`.
Parameters
----------
X : array-like
The features at which to predict
alpha : float
The significance level to use for the interval
Returns
-------
array, shape (2, n_samples)
Lower and upper bounds for the confidence interval at each sample point
"""
if self.fit_intercept:
X = add_constant(X, has_constant='add')
# NOTE: we use `obs = False` to get a confidence, rather than prediction, interval
preds = self.results.get_prediction(X).conf_int(alpha=alpha, obs=False)
# statsmodels uses the last dimension instead of the first to store the confidence intervals,
# so we need to transpose the result
return transpose(preds)
def coef__interval(self, alpha):
if self.fit_intercept:
return transpose(self.results.conf_int(alpha=alpha)[1:])
else:
return transpose(self.results.conf_int(alpha=alpha))