File size: 8,335 Bytes
7885a28 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 |
"""
Test extension array for storing nested data in a pandas container.
The JSONArray stores lists of dictionaries. The storage mechanism is a list,
not an ndarray.
Note
----
We currently store lists of UserDicts. Pandas has a few places
internally that specifically check for dicts, and does non-scalar things
in that case. We *want* the dictionaries to be treated as scalars, so we
hack around pandas by using UserDicts.
"""
from __future__ import annotations
from collections import (
UserDict,
abc,
)
import itertools
import numbers
import string
import sys
from typing import (
TYPE_CHECKING,
Any,
)
import numpy as np
from pandas.core.dtypes.cast import construct_1d_object_array_from_listlike
from pandas.core.dtypes.common import (
is_bool_dtype,
is_list_like,
pandas_dtype,
)
import pandas as pd
from pandas.api.extensions import (
ExtensionArray,
ExtensionDtype,
)
from pandas.core.indexers import unpack_tuple_and_ellipses
if TYPE_CHECKING:
from collections.abc import Mapping
from pandas._typing import type_t
class JSONDtype(ExtensionDtype):
type = abc.Mapping
name = "json"
na_value: Mapping[str, Any] = UserDict()
@classmethod
def construct_array_type(cls) -> type_t[JSONArray]:
"""
Return the array type associated with this dtype.
Returns
-------
type
"""
return JSONArray
class JSONArray(ExtensionArray):
dtype = JSONDtype()
__array_priority__ = 1000
def __init__(self, values, dtype=None, copy=False) -> None:
for val in values:
if not isinstance(val, self.dtype.type):
raise TypeError("All values must be of type " + str(self.dtype.type))
self.data = values
# Some aliases for common attribute names to ensure pandas supports
# these
self._items = self._data = self.data
# those aliases are currently not working due to assumptions
# in internal code (GH-20735)
# self._values = self.values = self.data
@classmethod
def _from_sequence(cls, scalars, *, dtype=None, copy=False):
return cls(scalars)
@classmethod
def _from_factorized(cls, values, original):
return cls([UserDict(x) for x in values if x != ()])
def __getitem__(self, item):
if isinstance(item, tuple):
item = unpack_tuple_and_ellipses(item)
if isinstance(item, numbers.Integral):
return self.data[item]
elif isinstance(item, slice) and item == slice(None):
# Make sure we get a view
return type(self)(self.data)
elif isinstance(item, slice):
# slice
return type(self)(self.data[item])
elif not is_list_like(item):
# e.g. "foo" or 2.5
# exception message copied from numpy
raise IndexError(
r"only integers, slices (`:`), ellipsis (`...`), numpy.newaxis "
r"(`None`) and integer or boolean arrays are valid indices"
)
else:
item = pd.api.indexers.check_array_indexer(self, item)
if is_bool_dtype(item.dtype):
return type(self)._from_sequence(
[x for x, m in zip(self, item) if m], dtype=self.dtype
)
# integer
return type(self)([self.data[i] for i in item])
def __setitem__(self, key, value) -> None:
if isinstance(key, numbers.Integral):
self.data[key] = value
else:
if not isinstance(value, (type(self), abc.Sequence)):
# broadcast value
value = itertools.cycle([value])
if isinstance(key, np.ndarray) and key.dtype == "bool":
# masking
for i, (k, v) in enumerate(zip(key, value)):
if k:
assert isinstance(v, self.dtype.type)
self.data[i] = v
else:
for k, v in zip(key, value):
assert isinstance(v, self.dtype.type)
self.data[k] = v
def __len__(self) -> int:
return len(self.data)
def __eq__(self, other):
return NotImplemented
def __ne__(self, other):
return NotImplemented
def __array__(self, dtype=None, copy=None):
if dtype is None:
dtype = object
if dtype == object:
# on py38 builds it looks like numpy is inferring to a non-1D array
return construct_1d_object_array_from_listlike(list(self))
return np.asarray(self.data, dtype=dtype)
@property
def nbytes(self) -> int:
return sys.getsizeof(self.data)
def isna(self):
return np.array([x == self.dtype.na_value for x in self.data], dtype=bool)
def take(self, indexer, allow_fill=False, fill_value=None):
# re-implement here, since NumPy has trouble setting
# sized objects like UserDicts into scalar slots of
# an ndarary.
indexer = np.asarray(indexer)
msg = (
"Index is out of bounds or cannot do a "
"non-empty take from an empty array."
)
if allow_fill:
if fill_value is None:
fill_value = self.dtype.na_value
# bounds check
if (indexer < -1).any():
raise ValueError
try:
output = [
self.data[loc] if loc != -1 else fill_value for loc in indexer
]
except IndexError as err:
raise IndexError(msg) from err
else:
try:
output = [self.data[loc] for loc in indexer]
except IndexError as err:
raise IndexError(msg) from err
return type(self)._from_sequence(output, dtype=self.dtype)
def copy(self):
return type(self)(self.data[:])
def astype(self, dtype, copy=True):
# NumPy has issues when all the dicts are the same length.
# np.array([UserDict(...), UserDict(...)]) fails,
# but np.array([{...}, {...}]) works, so cast.
from pandas.core.arrays.string_ import StringDtype
dtype = pandas_dtype(dtype)
# needed to add this check for the Series constructor
if isinstance(dtype, type(self.dtype)) and dtype == self.dtype:
if copy:
return self.copy()
return self
elif isinstance(dtype, StringDtype):
value = self.astype(str) # numpy doesn't like nested dicts
arr_cls = dtype.construct_array_type()
return arr_cls._from_sequence(value, dtype=dtype, copy=False)
elif not copy:
return np.asarray([dict(x) for x in self], dtype=dtype)
else:
return np.array([dict(x) for x in self], dtype=dtype, copy=copy)
def unique(self):
# Parent method doesn't work since np.array will try to infer
# a 2-dim object.
return type(self)([dict(x) for x in {tuple(d.items()) for d in self.data}])
@classmethod
def _concat_same_type(cls, to_concat):
data = list(itertools.chain.from_iterable(x.data for x in to_concat))
return cls(data)
def _values_for_factorize(self):
frozen = self._values_for_argsort()
if len(frozen) == 0:
# factorize_array expects 1-d array, this is a len-0 2-d array.
frozen = frozen.ravel()
return frozen, ()
def _values_for_argsort(self):
# Bypass NumPy's shape inference to get a (N,) array of tuples.
frozen = [tuple(x.items()) for x in self]
return construct_1d_object_array_from_listlike(frozen)
def _pad_or_backfill(self, *, method, limit=None, copy=True):
# GH#56616 - test EA method without limit_area argument
return super()._pad_or_backfill(method=method, limit=limit, copy=copy)
def make_data():
# TODO: Use a regular dict. See _NDFrameIndexer._setitem_with_indexer
rng = np.random.default_rng(2)
return [
UserDict(
[
(rng.choice(list(string.ascii_letters)), rng.integers(0, 100))
for _ in range(rng.integers(0, 10))
]
)
for _ in range(100)
]
|