def transform(self, X, y=None):
kwargs = self.kw_args if self.kw_args else {}
if isinstance(X, da.Array):
if self.validate:
X = check_array(X,
accept_dask_array=True,
accept_unknown_chunks=True)
XP = X.map_blocks(self.func,
dtype=X.dtype,
chunks=X.chunks,
**kwargs)
elif isinstance(X, dd.DataFrame):
if self.validate:
X = check_array(X,
accept_dask_dataframe=True,
preserve_pandas_dataframe=True)
XP = X.map_partitions(self.func, **kwargs)
elif isinstance(X, pd.DataFrame):
if self.validate:
X = check_array(X,
accept_dask_array=False,
preserve_pandas_dataframe=True)
XP = self.func(X, **kwargs)
else:
if self.validate:
X = check_array(X, accept_dask_array=False)
XP = self.func(X, **kwargs)
return XP
def transform(self, X, y=None):
if isinstance(X, da.Array):
n_cols = len(self._transformer.get_feature_names())
X = check_array(X,
accept_multiple_blocks=False,
accept_unknown_chunks=True)
chunks = (X.chunks[0], n_cols)
XP = X.map_blocks(self._transformer.transform,
dtype=X.dtype,
chunks=chunks)
elif isinstance(X, pd.DataFrame):
XP = X.pipe(self._transformer.transform)
if self.preserve_dataframe:
columns = self._transformer.get_feature_names(X.columns)
XP = pd.DataFrame(data=XP, columns=columns, index=X.index)
elif isinstance(X, dd.DataFrame):
XP = X.map_partitions(self._transformer.transform)
if self.preserve_dataframe:
columns = self._transformer.get_feature_names(X.columns)
XP = dd.from_dask_array(XP, columns, X.index)
else:
# typically X is instance of np.ndarray
XP = self._transformer.transform(X)
return XP
def _fit(self, X, handle_unknown="error"):
X = check_array(X,
accept_dask_dataframe=True,
dtype=None,
preserve_pandas_dataframe=True)
_, n_features = X.shape
if self.categories != "auto":
for cats in self.categories:
if not np.all(np.sort(cats) == np.array(cats)):
raise ValueError("Unsorted categories are not yet"
"supported")
if len(self.categories) != n_features:
raise ValueError("Shape mismatch: if n_values is an array,"
" it has to be of shape (n_features,).")
self.categories_ = []
self.dtypes_ = []
for i in range(n_features):
Xi = X[:, i]
if self.categories == "auto":
cats = _encode(Xi)
else:
cats = np.array(self.categories[i], dtype=X.dtype)
self.categories_.append(cats)
self.dtypes_.append(None)
self.categories_ = dask.compute(self.categories_)[0]
def check_array(array,
accept_sparse=False,
*,
accept_large_sparse=True,
dtype="numeric",
order=None,
copy=False,
force_all_finite=True,
ensure_2d=True,
allow_nd=False,
ensure_min_samples=1,
ensure_min_features=1,
estimator=None,
distributed=False,
chunks="16MB"):
_X = dmlu.check_array(array,
accept_sparse=accept_sparse,
accept_large_sparse=accept_large_sparse,
dtype=dtype,
order=order,
copy=copy,
force_all_finite=force_all_finite,
ensure_2d=ensure_2d,
allow_nd=allow_nd,
ensure_min_samples=ensure_min_samples,
ensure_min_features=ensure_min_features,
estimator=estimator)
if distributed:
return da.from_array(_X, chunks=chunks)
else:
return _X
def split(self, X, y=None, groups=None):
X = check_array(X)
for i in range(self.n_splits):
if self.blockwise:
yield self._split_blockwise(X)
else:
yield self._split(X)
def split(self, X, y=None, groups=None):
X = check_array(X)
rng = check_random_state(self.random_state)
for i in range(self.n_splits):
seeds = draw_seed(rng, 0, 2 ** 32 - 1, size=len(X.chunks[0]), dtype="uint")
if self.blockwise:
yield self._split_blockwise(X, seeds)
else:
yield self._split(X)
def split(self, X, y=None, groups=None):
X = check_array(X, ensure_2d=False, allow_nd=True)
rng = check_random_state(self.random_state)
for i in range(self.n_splits):
seeds = draw_seed(rng,
0,
_I4MAX,
size=len(X.chunks[0]),
dtype="uint")
if self.blockwise:
yield self._split_blockwise(X, seeds)
else:
yield self._split(X)
def split(self, X, y=None, groups=None):
X = check_array(X)
n_samples = X.shape[0]
n_splits = self.n_splits
fold_sizes = np.full(n_splits, n_samples // n_splits, dtype=np.int)
fold_sizes[:n_samples % n_splits] += 1
chunks = X.chunks[0]
seeds = [None] * len(chunks)
if self.shuffle:
rng = check_random_state(self.random_state)
seeds = draw_seed(rng, 0, _I4MAX, size=len(chunks), dtype="uint")
test_current = 0
for fold_size in fold_sizes:
test_start, test_stop = test_current, test_current + fold_size
yield self._split(test_start, test_stop, n_samples, chunks, seeds)
test_current = test_stop
def _transform_new(self, X):
X = check_array(X,
accept_dask_dataframe=True,
dtype=None,
preserve_pandas_dataframe=True)
_, n_features = X.shape
# We encode each column independently, as they have different categories.
Xs = [
_encode_dask_array(
X[:, i],
uniques=self.categories_[i],
encode=True,
onehot_dtype=self.dtype,
)[1] for i in range(n_features)
]
X = da.concatenate(Xs, axis=1)
if not self.sparse:
X = X.map_blocks(lambda x: x.toarray(), dtype=self.dtype)
return X.rechunk({1: X.shape[1]})
def _check_array(self, X, *args, **kwargs):
X = check_array(X, accept_dask_dataframe=True, **kwargs)
return X
def test_check_array_1d():
arr = da.random.uniform(size=(10, ), chunks=5)
check_array(arr, ensure_2d=False)
def test_check_array_raises():
X = da.random.uniform(size=(10, 5), chunks=2)
with pytest.raises(TypeError) as m:
check_array(X)
assert m.match("Chunking is only allowed on the first axis.")
def _check_array(self, X: Union[ArrayLike, DataFrameType], *args: Any,
**kwargs: Any) -> Union[ArrayLike, DataFrameType]:
X = check_array(X, accept_dask_dataframe=True, **kwargs)
return X
