Shortcuts on this page
r m x toggle line displays
j k next/prev highlighted chunk
0 (zero) top of page
1 (one) first highlighted chunk
1"""
2Core eval alignment algorithms.
3"""
4from __future__ import annotations
6from functools import (
7 partial,
8 wraps,
9)
10from typing import (
11 TYPE_CHECKING,
12 Callable,
13)
14import warnings
16import numpy as np
18from pandas.errors import PerformanceWarning
19from pandas.util._exceptions import find_stack_level
21from pandas.core.dtypes.generic import (
22 ABCDataFrame,
23 ABCSeries,
24)
26from pandas.core.base import PandasObject
27import pandas.core.common as com
28from pandas.core.computation.common import result_type_many
30if TYPE_CHECKING:
31 from collections.abc import Sequence
33 from pandas._typing import F
35 from pandas.core.generic import NDFrame
36 from pandas.core.indexes.api import Index
39def _align_core_single_unary_op(
40 term,
41) -> tuple[partial | type[NDFrame], dict[str, Index] | None]:
42 typ: partial | type[NDFrame]
43 axes: dict[str, Index] | None = None
45 if isinstance(term.value, np.ndarray):
46 typ = partial(np.asanyarray, dtype=term.value.dtype)
47 else:
48 typ = type(term.value)
49 if hasattr(term.value, "axes"):
50 axes = _zip_axes_from_type(typ, term.value.axes)
52 return typ, axes
55def _zip_axes_from_type(
56 typ: type[NDFrame], new_axes: Sequence[Index]
57) -> dict[str, Index]:
58 return {name: new_axes[i] for i, name in enumerate(typ._AXIS_ORDERS)}
61def _any_pandas_objects(terms) -> bool:
62 """
63 Check a sequence of terms for instances of PandasObject.
64 """
65 return any(isinstance(term.value, PandasObject) for term in terms)
68def _filter_special_cases(f) -> Callable[[F], F]:
69 @wraps(f)
70 def wrapper(terms):
71 # single unary operand
72 if len(terms) == 1:
73 return _align_core_single_unary_op(terms[0])
75 term_values = (term.value for term in terms)
77 # we don't have any pandas objects
78 if not _any_pandas_objects(terms):
79 return result_type_many(*term_values), None
81 return f(terms)
83 return wrapper
86@_filter_special_cases
87def _align_core(terms):
88 term_index = [i for i, term in enumerate(terms) if hasattr(term.value, "axes")]
89 term_dims = [terms[i].value.ndim for i in term_index]
91 from pandas import Series
93 ndims = Series(dict(zip(term_index, term_dims)))
95 # initial axes are the axes of the largest-axis'd term
96 biggest = terms[ndims.idxmax()].value
97 typ = biggest._constructor
98 axes = biggest.axes
99 naxes = len(axes)
100 gt_than_one_axis = naxes > 1
102 for value in (terms[i].value for i in term_index):
103 is_series = isinstance(value, ABCSeries)
104 is_series_and_gt_one_axis = is_series and gt_than_one_axis
106 for axis, items in enumerate(value.axes):
107 if is_series_and_gt_one_axis:
108 ax, itm = naxes - 1, value.index
109 else:
110 ax, itm = axis, items
112 if not axes[ax].is_(itm):
113 axes[ax] = axes[ax].union(itm)
115 for i, ndim in ndims.items():
116 for axis, items in zip(range(ndim), axes):
117 ti = terms[i].value
119 if hasattr(ti, "reindex"):
120 transpose = isinstance(ti, ABCSeries) and naxes > 1
121 reindexer = axes[naxes - 1] if transpose else items
123 term_axis_size = len(ti.axes[axis])
124 reindexer_size = len(reindexer)
126 ordm = np.log10(max(1, abs(reindexer_size - term_axis_size)))
127 if ordm >= 1 and reindexer_size >= 10000:
128 w = (
129 f"Alignment difference on axis {axis} is larger "
130 f"than an order of magnitude on term {repr(terms[i].name)}, "
131 f"by more than {ordm:.4g}; performance may suffer."
132 )
133 warnings.warn(
134 w, category=PerformanceWarning, stacklevel=find_stack_level()
135 )
137 obj = ti.reindex(reindexer, axis=axis, copy=False)
138 terms[i].update(obj)
140 terms[i].update(terms[i].value.values)
142 return typ, _zip_axes_from_type(typ, axes)
145def align_terms(terms):
146 """
147 Align a set of terms.
148 """
149 try:
150 # flatten the parse tree (a nested list, really)
151 terms = list(com.flatten(terms))
152 except TypeError:
153 # can't iterate so it must just be a constant or single variable
154 if isinstance(terms.value, (ABCSeries, ABCDataFrame)):
155 typ = type(terms.value)
156 return typ, _zip_axes_from_type(typ, terms.value.axes)
157 return np.result_type(terms.type), None
159 # if all resolved variables are numeric scalars
160 if all(term.is_scalar for term in terms):
161 return result_type_many(*(term.value for term in terms)).type, None
163 # perform the main alignment
164 typ, axes = _align_core(terms)
165 return typ, axes
168def reconstruct_object(typ, obj, axes, dtype):
169 """
170 Reconstruct an object given its type, raw value, and possibly empty
171 (None) axes.
173 Parameters
174 ----------
175 typ : object
176 A type
177 obj : object
178 The value to use in the type constructor
179 axes : dict
180 The axes to use to construct the resulting pandas object
182 Returns
183 -------
184 ret : typ
185 An object of type ``typ`` with the value `obj` and possible axes
186 `axes`.
187 """
188 try:
189 typ = typ.type
190 except AttributeError:
191 pass
193 res_t = np.result_type(obj.dtype, dtype)
195 if not isinstance(typ, partial) and issubclass(typ, PandasObject):
196 return typ(obj, dtype=res_t, **axes)
198 # special case for pathological things like ~True/~False
199 if hasattr(res_t, "type") and typ == np.bool_ and res_t != np.bool_:
200 ret_value = res_t.type(obj)
201 else:
202 ret_value = typ(obj).astype(res_t)
203 # The condition is to distinguish 0-dim array (returned in case of
204 # scalar) and 1 element array
205 # e.g. np.array(0) and np.array([0])
206 if (
207 len(obj.shape) == 1
208 and len(obj) == 1
209 and not isinstance(ret_value, np.ndarray)
210 ):
211 ret_value = np.array([ret_value]).astype(res_t)
213 return ret_value