Coverage for /pythoncovmergedfiles/medio/medio/usr/local/lib/python3.11/site-packages/pandas/core/arrays/numeric.py: 24%

1from __future__ import annotations 

2 

3import numbers 

4from typing import ( 

5 TYPE_CHECKING, 

6 Any, 

7 Self, 

8) 

9 

10import numpy as np 

11 

12from pandas._config import is_nan_na 

13 

14from pandas._libs import ( 

15 lib, 

16 missing as libmissing, 

17) 

18from pandas.errors import AbstractMethodError 

19from pandas.util._decorators import cache_readonly 

20 

21from pandas.core.dtypes.common import ( 

22 is_integer_dtype, 

23 is_string_dtype, 

24 pandas_dtype, 

25) 

26 

27from pandas.core.arrays.masked import ( 

28 BaseMaskedArray, 

29 BaseMaskedDtype, 

30) 

31 

32if TYPE_CHECKING: 

33 from collections.abc import ( 

34 Callable, 

35 Mapping, 

36 ) 

37 

38 import pyarrow 

39 

40 from pandas._typing import ( 

41 DtypeObj, 

42 npt, 

43 ) 

44 

45 from pandas.core.dtypes.dtypes import ExtensionDtype 

46 

47 

48class NumericDtype(BaseMaskedDtype): 

49 _default_np_dtype: np.dtype 

50 _checker: Callable[[Any], bool] # is_foo_dtype 

51 

52 def __repr__(self) -> str: 

53 return f"{self.name}Dtype()" 

54 

55 @cache_readonly 

56 def is_signed_integer(self) -> bool: 

57 return self.kind == "i" 

58 

59 @cache_readonly 

60 def is_unsigned_integer(self) -> bool: 

61 return self.kind == "u" 

62 

63 @property 

64 def _is_numeric(self) -> bool: 

65 return True 

66 

67 def __from_arrow__( 

68 self, array: pyarrow.Array | pyarrow.ChunkedArray 

69 ) -> BaseMaskedArray: 

70 """ 

71 Construct IntegerArray/FloatingArray from pyarrow Array/ChunkedArray. 

72 """ 

73 import pyarrow 

74 

75 from pandas.core.arrays.arrow._arrow_utils import ( 

76 pyarrow_array_to_numpy_and_mask, 

77 ) 

78 

79 array_class = self.construct_array_type() 

80 

81 pyarrow_type = pyarrow.from_numpy_dtype(self.type) 

82 if not array.type.equals(pyarrow_type) and not pyarrow.types.is_null( 

83 array.type 

84 ): 

85 # test_from_arrow_type_error raise for string, but allow 

86 # through itemsize conversion GH#31896 

87 rt_dtype = pandas_dtype(array.type.to_pandas_dtype()) 

88 if rt_dtype.kind not in "iuf": 

89 # Could allow "c" or potentially disallow float<->int conversion, 

90 # but at the moment we specifically test that uint<->int works 

91 raise TypeError( 

92 f"Expected array of {self} type, got {array.type} instead" 

93 ) 

94 

95 array = array.cast(pyarrow_type) 

96 

97 if isinstance(array, pyarrow.ChunkedArray): 

98 array = array.combine_chunks() 

99 

100 data, mask = pyarrow_array_to_numpy_and_mask(array, dtype=self.numpy_dtype) 

101 if data.dtype.kind == "f" and is_nan_na(): 

102 mask[np.isnan(data)] = False 

103 return array_class(data.copy(), ~mask, copy=False) 

104 

105 @classmethod 

106 def _get_dtype_mapping(cls) -> Mapping[np.dtype, NumericDtype]: 

107 raise AbstractMethodError(cls) 

108 

109 @classmethod 

110 def _standardize_dtype(cls, dtype: NumericDtype | str | np.dtype) -> NumericDtype: 

111 """ 

112 Convert a string representation or a numpy dtype to NumericDtype. 

113 """ 

114 if isinstance(dtype, str) and (dtype.startswith(("Int", "UInt", "Float"))): 

115 # Avoid DeprecationWarning from NumPy about np.dtype("Int64") 

116 # https://github.com/numpy/numpy/pull/7476 

117 dtype = dtype.lower() 

118 

119 if not isinstance(dtype, NumericDtype): 

120 mapping = cls._get_dtype_mapping() 

121 try: 

122 dtype = mapping[np.dtype(dtype)] 

123 except KeyError as err: 

124 raise ValueError(f"invalid dtype specified {dtype}") from err 

125 return dtype 

126 

127 @classmethod 

128 def _safe_cast(cls, values: np.ndarray, dtype: np.dtype, copy: bool) -> np.ndarray: 

129 """ 

130 Safely cast the values to the given dtype. 

131 

132 "safe" in this context means the casting is lossless. 

133 """ 

134 raise AbstractMethodError(cls) 

135 

136 

137def _coerce_to_data_and_mask(values, dtype, copy: bool, dtype_cls: type[NumericDtype]): 

138 checker = dtype_cls._checker 

139 default_dtype = dtype_cls._default_np_dtype 

140 

141 mask = None 

142 inferred_type = None 

143 

144 if dtype is None and hasattr(values, "dtype"): 

145 if checker(values.dtype): 

146 dtype = values.dtype 

147 

148 if dtype is not None: 

149 dtype = dtype_cls._standardize_dtype(dtype) 

150 

151 cls = dtype_cls().construct_array_type() 

152 if isinstance(values, cls): 

153 values, mask = values._data, values._mask 

154 if dtype is not None: 

155 values = values.astype(dtype.numpy_dtype, copy=False) 

156 

157 if copy: 

158 values = values.copy() 

159 mask = mask.copy() 

160 return values, mask 

161 

162 original = values 

163 if not copy: 

164 values = np.asarray(values) 

165 else: 

166 values = np.array(values, copy=copy) 

167 inferred_type = None 

168 if values.dtype == object or is_string_dtype(values.dtype): 

169 inferred_type = lib.infer_dtype(values, skipna=True) 

170 if inferred_type == "boolean" and dtype is None: 

171 # object dtype array of bools 

172 name = dtype_cls.__name__.strip("_") 

173 raise TypeError(f"{values.dtype} cannot be converted to {name}") 

174 

175 elif values.dtype.kind == "b" and checker(dtype): 

176 # fastpath 

177 mask = np.zeros(len(values), dtype=np.bool_) 

178 if not copy: 

179 values = np.asarray(values, dtype=default_dtype) 

180 else: 

181 values = np.array(values, dtype=default_dtype, copy=copy) 

182 

183 elif values.dtype.kind not in "iuf": 

184 name = dtype_cls.__name__.strip("_") 

185 raise TypeError(f"{values.dtype} cannot be converted to {name}") 

186 

187 if values.ndim != 1: 

188 raise TypeError("values must be a 1D list-like") 

189 

190 if mask is None: 

191 if values.dtype.kind in "iu": 

192 # fastpath 

193 mask = np.zeros(len(values), dtype=np.bool_) 

194 elif values.dtype.kind == "f": 

195 # np.isnan is faster than is_numeric_na() for floats 

196 # github issue: #60066 

197 if is_nan_na(): 

198 mask = np.isnan(values) 

199 else: 

200 mask = np.zeros(len(values), dtype=np.bool_) 

201 if dtype_cls.__name__.strip("_").startswith(("I", "U")): 

202 wrong = np.isnan(values) 

203 if wrong.any(): 

204 raise ValueError("Cannot cast NaN value to Integer dtype.") 

205 elif is_nan_na(): 

206 mask = libmissing.is_numeric_na(values) 

207 else: 

208 # is_numeric_na will raise on non-numeric NAs 

209 libmissing.is_numeric_na(values) 

210 mask = libmissing.is_pdna_or_none(values) 

211 else: 

212 assert len(mask) == len(values) 

213 

214 if mask.ndim != 1: 

215 raise TypeError("mask must be a 1D list-like") 

216 

217 # infer dtype if needed 

218 if dtype is None: 

219 dtype = default_dtype 

220 else: 

221 dtype = dtype.numpy_dtype 

222 

223 if is_integer_dtype(dtype) and values.dtype.kind == "f" and len(values) > 0: 

224 if mask.all(): 

225 values = np.ones(values.shape, dtype=dtype) 

226 else: 

227 idx = np.nanargmax(values) 

228 if int(values[idx]) != original[idx]: 

229 # We have ints that lost precision during the cast. 

230 inferred_type = lib.infer_dtype(original, skipna=True) 

231 if ( 

232 inferred_type not in ["floating", "mixed-integer-float"] 

233 and not mask.any() 

234 ): 

235 values = np.asarray(original, dtype=dtype) 

236 else: 

237 values = np.asarray(original, dtype="object") 

238 

239 # we copy as need to coerce here 

240 if mask.any(): 

241 values = values.copy() 

242 values[mask] = dtype_cls._internal_fill_value 

243 if inferred_type in ("string", "unicode"): 

244 # casts from str are always safe since they raise 

245 # a ValueError if the str cannot be parsed into a float 

246 values = values.astype(dtype, copy=copy) 

247 else: 

248 values = dtype_cls._safe_cast(values, dtype, copy=False) 

249 return values, mask 

250 

251 

252class NumericArray(BaseMaskedArray): 

253 """ 

254 Base class for IntegerArray and FloatingArray. 

255 """ 

256 

257 _dtype_cls: type[NumericDtype] 

258 

259 def __init__( 

260 self, values: np.ndarray, mask: npt.NDArray[np.bool_], copy: bool = False 

261 ) -> None: 

262 checker = self._dtype_cls._checker 

263 if not (isinstance(values, np.ndarray) and checker(values.dtype)): 

264 descr = ( 

265 "floating" 

266 if self._dtype_cls.kind == "f" # type: ignore[comparison-overlap] 

267 else "integer" 

268 ) 

269 raise TypeError( 

270 f"values should be {descr} numpy array. Use " 

271 "the 'pd.array' function instead" 

272 ) 

273 if values.dtype == np.float16: 

274 # If we don't raise here, then accessing self.dtype would raise 

275 raise TypeError("FloatingArray does not support np.float16 dtype.") 

276 

277 # NB: if is_nan_na() is True 

278 # then caller is responsible for ensuring 

279 # assert mask[np.isnan(values)].all() 

280 

281 super().__init__(values, mask, copy=copy) 

282 

283 @cache_readonly 

284 def dtype(self) -> NumericDtype: 

285 mapping = self._dtype_cls._get_dtype_mapping() 

286 return mapping[self._data.dtype] 

287 

288 @classmethod 

289 def _coerce_to_array( 

290 cls, value, *, dtype: DtypeObj, copy: bool = False 

291 ) -> tuple[np.ndarray, np.ndarray]: 

292 dtype_cls = cls._dtype_cls 

293 values, mask = _coerce_to_data_and_mask(value, dtype, copy, dtype_cls) 

294 return values, mask 

295 

296 @classmethod 

297 def _from_sequence_of_strings( 

298 cls, strings, *, dtype: ExtensionDtype, copy: bool = False 

299 ) -> Self: 

300 from pandas.core.tools.numeric import to_numeric 

301 

302 scalars = to_numeric(strings, errors="raise", dtype_backend="numpy_nullable") 

303 return cls._from_sequence(scalars, dtype=dtype, copy=copy) 

304 

305 _HANDLED_TYPES = (np.ndarray, numbers.Number)