File: views.py

package info (click to toggle)
python-anndata 0.7.5%2Bds-3
  • links: PTS, VCS
  • area: main
  • in suites: bullseye
  • size: 628 kB
  • sloc: python: 7,779; makefile: 8
file content (191 lines) | stat: -rw-r--r-- 5,127 bytes parent folder | download
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
from contextlib import contextmanager
from copy import deepcopy
from functools import reduce, singledispatch, wraps
from typing import Any, KeysView, Optional, Sequence, Tuple

import numpy as np
import pandas as pd
from pandas.api.types import is_bool_dtype
from scipy import sparse

from .access import ElementRef
from ..logging import anndata_logger as logger
from ..compat import ZappyArray


class _SetItemMixin:
    """\
    Class which (when values are being set) lets their parent AnnData view know,
    so it can make a copy of itself.
    This implements copy-on-modify semantics for views of AnnData objects.
    """

    def __setitem__(self, idx: Any, value: Any):
        if self._view_args is None:
            super().__setitem__(idx, value)
        else:
            logger.warning(
                f"Trying to set attribute `.{self._view_args.attrname}` of view, copying."
            )
            with self._update() as container:
                container[idx] = value

    @contextmanager
    def _update(self):
        adata_view, attr_name, keys = self._view_args
        new = adata_view.copy()
        attr = getattr(new, attr_name)
        container = reduce(lambda d, k: d[k], keys, attr)
        yield container
        adata_view._init_as_actual(new)


class _ViewMixin(_SetItemMixin):
    def __init__(
        self,
        *args,
        view_args: Tuple["anndata.AnnData", str, Tuple[str, ...]] = None,
        **kwargs,
    ):
        if view_args is not None:
            view_args = ElementRef(*view_args)
        self._view_args = view_args
        super().__init__(*args, **kwargs)

    # TODO: This makes `deepcopy(obj)` return `obj._view_args.parent._adata_ref`, fix it
    def __deepcopy__(self, memo):
        parent, attrname, keys = self._view_args
        return deepcopy(getattr(parent._adata_ref, attrname))


class ArrayView(_SetItemMixin, np.ndarray):
    def __new__(
        cls,
        input_array: Sequence[Any],
        view_args: Tuple["anndata.AnnData", str, Tuple[str, ...]] = None,
    ):
        arr = np.asanyarray(input_array).view(cls)

        if view_args is not None:
            view_args = ElementRef(*view_args)
        arr._view_args = view_args
        return arr

    def __array_finalize__(self, obj: Optional[np.ndarray]):
        if obj is not None:
            self._view_args = getattr(obj, "_view_args", None)

    def keys(self) -> KeysView[str]:
        # it’s a structured array
        return self.dtype.names

    def copy(self, order: str = "C") -> np.ndarray:
        # we want a conventional array
        return np.array(self)

    def toarray(self) -> np.ndarray:
        return self.copy()


# Unlike array views, SparseCSRView and SparseCSCView
# do not propagate through subsetting
class SparseCSRView(_ViewMixin, sparse.csr_matrix):
    pass


class SparseCSCView(_ViewMixin, sparse.csc_matrix):
    pass


class DictView(_ViewMixin, dict):
    pass


class DataFrameView(_ViewMixin, pd.DataFrame):
    _metadata = ["_view_args"]

    @wraps(pd.DataFrame.drop)
    def drop(self, *args, inplace: bool = False, **kw):
        if not inplace:
            return self.copy().drop(*args, **kw)
        with self._update() as df:
            df.drop(*args, inplace=True, **kw)


@singledispatch
def as_view(obj, view_args):
    raise NotImplementedError(f"No view type has been registered for {type(obj)}")


@as_view.register(np.ndarray)
def as_view_array(array, view_args):
    return ArrayView(array, view_args=view_args)


@as_view.register(pd.DataFrame)
def as_view_df(df, view_args):
    return DataFrameView(df, view_args=view_args)


@as_view.register(sparse.csr_matrix)
def as_view_csr(mtx, view_args):
    return SparseCSRView(mtx, view_args=view_args)


@as_view.register(sparse.csc_matrix)
def as_view_csc(mtx, view_args):
    return SparseCSCView(mtx, view_args=view_args)


@as_view.register(dict)
def as_view_dict(d, view_args):
    return DictView(d, view_args=view_args)


@as_view.register(ZappyArray)
def as_view_zappy(z, view_args):
    # Previous code says ZappyArray works as view,
    # but as far as I can tell they’re immutable.
    return z


def _resolve_idxs(old, new, adata):
    t = tuple(_resolve_idx(old[i], new[i], adata.shape[i]) for i in (0, 1))
    return t


@singledispatch
def _resolve_idx(old, new, l):
    return old[new]


@_resolve_idx.register(np.ndarray)
def _resolve_idx_ndarray(old, new, l):
    if is_bool_dtype(old):
        old = np.where(old)[0]
    return old[new]


@_resolve_idx.register(np.integer)
@_resolve_idx.register(int)
def _resolve_idx_scalar(old, new, l):
    return np.array([old])[new]


@_resolve_idx.register(slice)
def _resolve_idx_slice(old, new, l):
    if isinstance(new, slice):
        return _resolve_idx_slice_slice(old, new, l)
    else:
        return np.arange(*old.indices(l))[new]


def _resolve_idx_slice_slice(old, new, l):
    r = range(*old.indices(l))[new]
    # Convert back to slice
    start, stop, step = r.start, r.stop, r.step
    if len(r) == 0:
        stop = start
    elif stop < 0:
        stop = None
    return slice(start, stop, step)