Initial CSC/ CSR classes

sparse/_compressed/__init__.py

@@ -1,2 +1,2 @@
-from .compressed import GCXS
+from .compressed import GCXS, CSC, CSR
 from .common import stack, concatenate

sparse/_compressed/compressed.py

@@ -5,6 +5,8 @@
 from functools import reduce
 from collections.abc import Iterable
 import scipy.sparse as ss
+from scipy.sparse import compressed
+from typing import Tuple
 
 from .._sparse_array import SparseArray, _reduce_super_ufunc
 from .._coo.common import linear_loc
@@ -136,6 +138,9 @@ def __init__(
         idx_dtype=None,
     ):
 
+        if isinstance(arg, ss.spmatrix):
+            arg = self.from_scipy_sparse(arg)
+
         if isinstance(arg, np.ndarray):
             (arg, shape, compressed_axes, fill_value) = _from_coo(
                 COO(arg), compressed_axes
@@ -146,6 +151,16 @@ def __init__(
                 arg, compressed_axes, idx_dtype
             )
 
+        elif isinstance(arg, GCXS):
+            if compressed_axes is not None and arg.compressed_axes != compressed_axes:
+                arg = arg.change_compressed_axes(self.compressed_axes)
+            (arg, shape, compressed_axes, fill_value) = (
+                (arg.data, arg.indices, arg.indptr),
+                arg.shape,
+                arg.compressed_axes,
+                arg.fill_value,
+            )
+
         if shape is None:
             raise ValueError("missing `shape` argument")
 
@@ -160,6 +175,7 @@ def __init__(
             raise ValueError("data must be a scalar or 1-dimensional.")
 
         self.shape = shape
+
         self.compressed_axes = (
             tuple(compressed_axes) if isinstance(compressed_axes, Iterable) else None
         )
@@ -440,7 +456,7 @@ def todense(self):
 
     def todok(self):
 
-        from ..dok import DOK
+        from .. import DOK
 
         return DOK.from_coo(self.tocoo())  # probably a temporary solution
 
@@ -496,6 +512,10 @@ def asformat(self, format, compressed_axes=None):
             return self.tocoo()
         elif format == "dok":
             return self.todok()
+        elif format == "csr":
+            return CSR(self)
+        elif format == "csc":
+            return CSC(self)
         elif format == "gcxs":
             if compressed_axes is None:
                 compressed_axes = self.compressed_axes
@@ -817,3 +837,83 @@ def _prune(self):
             self.indptr = indptr
         else:
             self.indices = self.indices[mask]
+
+
+class Compressed2d(GCXS):
+    def __init__(self, arg, shape=None, prune=False, fill_value=0):
+        if not hasattr(arg, "shape") and shape is None:
+            raise ValueError("missing `shape` argument")
+        if shape is not None and hasattr(arg, "shape"):
+            raise NotImplementedError("Cannot change shape in constructor")
+        nd = len(shape if shape is not None else arg.shape)
+        if nd != 2:
+            raise ValueError(f"{type(self).__name__} must be 2-d, passed {nd}-d shape.")
+
+        super().__init__(
+            arg,
+            shape=shape,
+            compressed_axes=self.compressed_axes,
+            prune=prune,
+            fill_value=fill_value,
+        )
+
+    def __str__(self):
+        return "<{}: shape={}, dtype={}, nnz={}, fill_value={}>".format(
+            type(self).__name__,
+            self.shape,
+            self.dtype,
+            self.nnz,
+            self.fill_value,
+        )
+
+    __repr__ = __str__
+
+    @property
+    def ndim(self) -> int:
+        return 2
+
+
+class CSR(Compressed2d):
+    @classmethod
+    def from_scipy_sparse(cls, x):
+        x = x.asformat("csr", copy=False)
+        return cls((x.data, x.indices, x.indptr), shape=x.shape)
+
+    @property
+    def compressed_axes(self) -> int:
+        return (0,)
+
+    @compressed_axes.setter
+    def compressed_axes(self, val):
+        if val != self.compressed_axes:
+            raise ValueError()
+
+    def transpose(self, axes: None = None, copy: bool = False) -> "CSC":
+        if axes is not None:
+            raise ValueError()
+        if copy:
+            self = self.copy()
+        return CSC((self.data, self.indices, self.indptr), self.shape[::-1])
+
+
+class CSC(Compressed2d):
+    @classmethod
+    def from_scipy_sparse(cls, x):
+        x = x.asformat("csc", copy=False)
+        return cls((x.data, x.indices, x.indptr), shape=x.shape)
+
+    @property
+    def compressed_axes(self) -> int:
+        return (1,)
+
+    @compressed_axes.setter
+    def compressed_axes(self, val):
+        if val != self.compressed_axes:
+            raise ValueError()
+
+    def transpose(self, axes: None = None, copy: bool = False) -> CSR:
+        if axes is not None:
+            raise ValueError()
+        if copy:
+            self = self.copy()
+        return CSR((self.data, self.indices, self.indptr), self.shape[::-1])

sparse/tests/test_compressed_2d.py

@@ -0,0 +1,124 @@
+import numpy as np
+from numpy.core.numeric import indices
+import pytest
+import scipy.sparse
+from scipy.sparse import data
+from scipy.sparse.construct import random
+import scipy.stats
+
+import sparse
+from sparse import COO
+from sparse._compressed.compressed import GCXS, CSR, CSC
+from sparse._utils import assert_eq
+
+
+@pytest.fixture(scope="module", params=[CSR, CSC])
+def cls(request):
+    return request.param
+
+
+@pytest.fixture(scope="module", params=["f8", "f4", "i8", "i4"])
+def dtype(request):
+    return request.param
+
+
+@pytest.fixture(scope="module")
+def random_sparse(cls, dtype):
+    if np.issubdtype(dtype, np.integer):
+
+        def data_rvs(n):
+            return np.random.randint(-1000, 1000, n)
+
+    else:
+        data_rvs = None
+    return cls(sparse.random((20, 30), density=0.25, data_rvs=data_rvs).astype(dtype))
+
+
+@pytest.fixture(scope="module")
+def random_sparse_small(cls, dtype):
+    if np.issubdtype(dtype, np.integer):
+
+        def data_rvs(n):
+            return np.random.randint(-10, 10, n)
+
+    else:
+        data_rvs = None
+    return cls(
+        sparse.random((20, 30, 40), density=0.25, data_rvs=data_rvs).astype(dtype)
+    )
+
+
+def test_repr(random_sparse):
+    cls = type(random_sparse).__name__
+
+    str_repr = repr(random_sparse)
+    assert cls in str_repr
+
+
+def test_bad_constructor_input(cls):
+    with pytest.raises(ValueError, match=r".*shape.*"):
+        cls(arg="hello world")
+
+
+@pytest.mark.parametrize("n", [0, 1, 3])
+def test_bad_nd_input(cls, n):
+    a = np.ones(shape=tuple(5 for _ in range(n)))
+    with pytest.raises(ValueError, match=f"{n}-d"):
+        cls(a)
+
+
+@pytest.mark.parametrize("source_type", ["gcxs", "coo"])
+def test_from_sparse(cls, source_type):
+    gcxs = sparse.random((20, 30), density=0.25, format=source_type)
+    result = cls(gcxs)
+
+    assert_eq(result, gcxs)
+
+
+@pytest.mark.parametrize("scipy_type", ["coo", "csr", "csc", "lil"])
+@pytest.mark.parametrize("CLS", [CSR, CSC, GCXS])
+def test_from_scipy_sparse(scipy_type, CLS, dtype):
+    orig = scipy.sparse.random(20, 30, density=0.2, format=scipy_type, dtype=dtype)
+    ref = COO.from_scipy_sparse(orig)
+    result = CLS.from_scipy_sparse(orig)
+
+    assert_eq(ref, result)
+
+    result_via_init = CLS(orig)
+
+    assert_eq(ref, result_via_init)
+
+
+@pytest.mark.parametrize("cls_str", ["coo", "dok", "csr", "csc", "gcxs"])
+def test_to_sparse(cls_str, random_sparse):
+    result = random_sparse.asformat(cls_str)
+
+    assert_eq(random_sparse, result)
+
+
+@pytest.mark.parametrize("copy", [True, False])
+def test_transpose(random_sparse, copy):
+    from operator import is_, is_not
+
+    t = random_sparse.transpose(copy=copy)
+    tt = t.transpose(copy=copy)
+
+    # Check if a copy was made
+    if copy:
+        check = is_not
+    else:
+        check = is_
+
+    assert check(random_sparse.data, t.data)
+    assert check(random_sparse.indices, t.indices)
+    assert check(random_sparse.indptr, t.indptr)
+
+    assert random_sparse.shape == t.shape[::-1]
+
+    assert_eq(random_sparse, tt)
+    assert type(random_sparse) == type(tt)
+
+
+def test_transpose_error(random_sparse):
+    with pytest.raises(ValueError):
+        random_sparse.transpose(axes=1)