def test_unknown_shapes(fn, solver):
rng = sk_check_random_state(42)
X = rng.uniform(-1, 1, size=(10, 3))
df = pd.DataFrame(X)
ddf = dask.dataframe.from_pandas(df, npartitions=2)
pca = dd.PCA(n_components=2, svd_solver=solver)
fit_fn = getattr(pca, fn)
X = ddf.values
assert np.isnan(X.shape[0])
if solver == "auto":
with pytest.raises(ValueError, match="Cannot automatically choose PCA solver"):
fit_fn(X)
else:
X_hat = fit_fn(X)
assert hasattr(pca, "components_")
assert pca.n_components_ == 2
assert pca.n_features_ == 3
assert np.isnan(pca.n_samples_)
if fn == "fit_transform":
assert np.isnan(X_hat.shape[0])
assert X_hat.shape[1] == 2
def split_samples(X, y, fractions=[0.75, 0.25], random_state=None):
"""Split samples into training, test, and cross-validation sets
Parameters
----------
X, y : array_like
leading dimension n_samples
fraction : array_like
length n_splits. If the fractions do not add to 1, they will be
re-normalized.
random_state : None, int, or RandomState object
random seed, or random number generator
"""
X = np.asarray(X)
y = np.asarray(y)
if X.shape[0] != y.shape[0]:
raise ValueError("X and y should have the same leading dimension")
n_samples = X.shape[0]
fractions = np.asarray(fractions).ravel().cumsum()
fractions /= fractions[-1]
fractions *= n_samples
N = np.concatenate([[0], fractions.astype(int)])
N[-1] = n_samples # in case of roundoff errors
random_state = sk_check_random_state(random_state)
indices = np.arange(len(y))
random_state.shuffle(indices)
X_divisions = tuple(X[indices[N[i]:N[i + 1]]]
for i in range(len(fractions)))
y_divisions = tuple(y[indices[N[i]:N[i + 1]]]
for i in range(len(fractions)))
return X_divisions, y_divisions
def check_random_state(seed):
return sk_check_random_state(seed)