Coverage for /var/srv/projects/api.amasfac.comuna18.com/tmp/venv/lib/python3.9/site-packages/pandas/core/tools/numeric.py: 7%
84 statements
« prev ^ index » next coverage.py v6.4.4, created at 2023-07-17 14:22 -0600
« prev ^ index » next coverage.py v6.4.4, created at 2023-07-17 14:22 -0600
1from __future__ import annotations
3import numpy as np
5from pandas._libs import lib
6from pandas._typing import npt
8from pandas.core.dtypes.cast import maybe_downcast_numeric
9from pandas.core.dtypes.common import (
10 ensure_object,
11 is_datetime_or_timedelta_dtype,
12 is_decimal,
13 is_integer_dtype,
14 is_number,
15 is_numeric_dtype,
16 is_scalar,
17 needs_i8_conversion,
18)
19from pandas.core.dtypes.generic import (
20 ABCIndex,
21 ABCSeries,
22)
24import pandas as pd
25from pandas.core.arrays.numeric import NumericArray
28def to_numeric(arg, errors="raise", downcast=None):
29 """
30 Convert argument to a numeric type.
32 The default return dtype is `float64` or `int64`
33 depending on the data supplied. Use the `downcast` parameter
34 to obtain other dtypes.
36 Please note that precision loss may occur if really large numbers
37 are passed in. Due to the internal limitations of `ndarray`, if
38 numbers smaller than `-9223372036854775808` (np.iinfo(np.int64).min)
39 or larger than `18446744073709551615` (np.iinfo(np.uint64).max) are
40 passed in, it is very likely they will be converted to float so that
41 they can stored in an `ndarray`. These warnings apply similarly to
42 `Series` since it internally leverages `ndarray`.
44 Parameters
45 ----------
46 arg : scalar, list, tuple, 1-d array, or Series
47 Argument to be converted.
48 errors : {'ignore', 'raise', 'coerce'}, default 'raise'
49 - If 'raise', then invalid parsing will raise an exception.
50 - If 'coerce', then invalid parsing will be set as NaN.
51 - If 'ignore', then invalid parsing will return the input.
52 downcast : str, default None
53 Can be 'integer', 'signed', 'unsigned', or 'float'.
54 If not None, and if the data has been successfully cast to a
55 numerical dtype (or if the data was numeric to begin with),
56 downcast that resulting data to the smallest numerical dtype
57 possible according to the following rules:
59 - 'integer' or 'signed': smallest signed int dtype (min.: np.int8)
60 - 'unsigned': smallest unsigned int dtype (min.: np.uint8)
61 - 'float': smallest float dtype (min.: np.float32)
63 As this behaviour is separate from the core conversion to
64 numeric values, any errors raised during the downcasting
65 will be surfaced regardless of the value of the 'errors' input.
67 In addition, downcasting will only occur if the size
68 of the resulting data's dtype is strictly larger than
69 the dtype it is to be cast to, so if none of the dtypes
70 checked satisfy that specification, no downcasting will be
71 performed on the data.
73 Returns
74 -------
75 ret
76 Numeric if parsing succeeded.
77 Return type depends on input. Series if Series, otherwise ndarray.
79 See Also
80 --------
81 DataFrame.astype : Cast argument to a specified dtype.
82 to_datetime : Convert argument to datetime.
83 to_timedelta : Convert argument to timedelta.
84 numpy.ndarray.astype : Cast a numpy array to a specified type.
85 DataFrame.convert_dtypes : Convert dtypes.
87 Examples
88 --------
89 Take separate series and convert to numeric, coercing when told to
91 >>> s = pd.Series(['1.0', '2', -3])
92 >>> pd.to_numeric(s)
93 0 1.0
94 1 2.0
95 2 -3.0
96 dtype: float64
97 >>> pd.to_numeric(s, downcast='float')
98 0 1.0
99 1 2.0
100 2 -3.0
101 dtype: float32
102 >>> pd.to_numeric(s, downcast='signed')
103 0 1
104 1 2
105 2 -3
106 dtype: int8
107 >>> s = pd.Series(['apple', '1.0', '2', -3])
108 >>> pd.to_numeric(s, errors='ignore')
109 0 apple
110 1 1.0
111 2 2
112 3 -3
113 dtype: object
114 >>> pd.to_numeric(s, errors='coerce')
115 0 NaN
116 1 1.0
117 2 2.0
118 3 -3.0
119 dtype: float64
121 Downcasting of nullable integer and floating dtypes is supported:
123 >>> s = pd.Series([1, 2, 3], dtype="Int64")
124 >>> pd.to_numeric(s, downcast="integer")
125 0 1
126 1 2
127 2 3
128 dtype: Int8
129 >>> s = pd.Series([1.0, 2.1, 3.0], dtype="Float64")
130 >>> pd.to_numeric(s, downcast="float")
131 0 1.0
132 1 2.1
133 2 3.0
134 dtype: Float32
135 """
136 if downcast not in (None, "integer", "signed", "unsigned", "float"):
137 raise ValueError("invalid downcasting method provided")
139 if errors not in ("ignore", "raise", "coerce"):
140 raise ValueError("invalid error value specified")
142 is_series = False
143 is_index = False
144 is_scalars = False
146 if isinstance(arg, ABCSeries):
147 is_series = True
148 values = arg.values
149 elif isinstance(arg, ABCIndex):
150 is_index = True
151 if needs_i8_conversion(arg.dtype):
152 values = arg.asi8
153 else:
154 values = arg.values
155 elif isinstance(arg, (list, tuple)):
156 values = np.array(arg, dtype="O")
157 elif is_scalar(arg):
158 if is_decimal(arg):
159 return float(arg)
160 if is_number(arg):
161 return arg
162 is_scalars = True
163 values = np.array([arg], dtype="O")
164 elif getattr(arg, "ndim", 1) > 1:
165 raise TypeError("arg must be a list, tuple, 1-d array, or Series")
166 else:
167 values = arg
169 # GH33013: for IntegerArray & FloatingArray extract non-null values for casting
170 # save mask to reconstruct the full array after casting
171 mask: npt.NDArray[np.bool_] | None = None
172 if isinstance(values, NumericArray):
173 mask = values._mask
174 values = values._data[~mask]
176 values_dtype = getattr(values, "dtype", None)
177 if is_numeric_dtype(values_dtype):
178 pass
179 elif is_datetime_or_timedelta_dtype(values_dtype):
180 values = values.view(np.int64)
181 else:
182 values = ensure_object(values)
183 coerce_numeric = errors not in ("ignore", "raise")
184 try:
185 values, _ = lib.maybe_convert_numeric(
186 values, set(), coerce_numeric=coerce_numeric
187 )
188 except (ValueError, TypeError):
189 if errors == "raise":
190 raise
192 # attempt downcast only if the data has been successfully converted
193 # to a numerical dtype and if a downcast method has been specified
194 if downcast is not None and is_numeric_dtype(values.dtype):
195 typecodes: str | None = None
197 if downcast in ("integer", "signed"):
198 typecodes = np.typecodes["Integer"]
199 elif downcast == "unsigned" and (not len(values) or np.min(values) >= 0):
200 typecodes = np.typecodes["UnsignedInteger"]
201 elif downcast == "float":
202 typecodes = np.typecodes["Float"]
204 # pandas support goes only to np.float32,
205 # as float dtypes smaller than that are
206 # extremely rare and not well supported
207 float_32_char = np.dtype(np.float32).char
208 float_32_ind = typecodes.index(float_32_char)
209 typecodes = typecodes[float_32_ind:]
211 if typecodes is not None:
212 # from smallest to largest
213 for typecode in typecodes:
214 dtype = np.dtype(typecode)
215 if dtype.itemsize <= values.dtype.itemsize:
216 values = maybe_downcast_numeric(values, dtype)
218 # successful conversion
219 if values.dtype == dtype:
220 break
222 # GH33013: for IntegerArray & FloatingArray need to reconstruct masked array
223 if mask is not None:
224 data = np.zeros(mask.shape, dtype=values.dtype)
225 data[~mask] = values
227 from pandas.core.arrays import (
228 FloatingArray,
229 IntegerArray,
230 )
232 klass = IntegerArray if is_integer_dtype(data.dtype) else FloatingArray
233 values = klass(data, mask.copy())
235 if is_series:
236 return arg._constructor(values, index=arg.index, name=arg.name)
237 elif is_index:
238 # because we want to coerce to numeric if possible,
239 # do not use _shallow_copy
240 return pd.Index(values, name=arg.name)
241 elif is_scalars:
242 return values[0]
243 else:
244 return values