# Licensed under a 3-clause BSD style license - see LICENSE.rst
"""This module contains functions and methods that relate to the DataInfo class
which provides a container for informational attributes as well as summary info
methods.
A DataInfo object is attached to the Quantity, SkyCoord, and Time classes in
astropy. Here it allows those classes to be used in Tables and uniformly carry
table column attributes such as name, format, dtype, meta, and description.
"""
# Note: these functions and classes are tested extensively in astropy table
# tests via their use in providing mixin column info, and in
# astropy/tests/test_info for providing table and column info summary data.
import os
import re
import sys
import warnings
import weakref
from collections import OrderedDict
from contextlib import contextmanager
from copy import deepcopy
from functools import partial
from io import StringIO
import numpy as np
from . import metadata
__all__ = [
"data_info_factory",
"dtype_info_name",
"BaseColumnInfo",
"DataInfo",
"MixinInfo",
"ParentDtypeInfo",
]
# Tuple of filterwarnings kwargs to ignore when calling info
IGNORE_WARNINGS = (
dict(
category=RuntimeWarning,
message=(
"All-NaN|"
"Mean of empty slice|Degrees of freedom <= 0|"
"invalid value encountered in sqrt"
),
),
)
@contextmanager
def serialize_context_as(context):
"""Set context for serialization.
This will allow downstream code to understand the context in which a column
is being serialized. Objects like Time or SkyCoord will have different
default serialization representations depending on context.
Parameters
----------
context : str
Context name, e.g. 'fits', 'hdf5', 'parquet', 'ecsv', 'yaml'
"""
old_context = BaseColumnInfo._serialize_context
BaseColumnInfo._serialize_context = context
try:
yield
finally:
BaseColumnInfo._serialize_context = old_context
[docs]
def dtype_info_name(dtype):
"""Return a human-oriented string name of the ``dtype`` arg.
This can be use by astropy methods that present type information about
a data object.
The output is mostly equivalent to ``dtype.name`` which takes the form
<type_name>[B] where <type_name> is like ``int`` or ``bool`` and [B] is an
optional number of bits which gets included only for numeric types.
The output is shown below for ``bytes`` and ``str`` types, with <N> being
the number of characters. This representation corresponds to the Python
type that matches the dtype::
Numpy S<N> U<N>
Python bytes<N> str<N>
Parameters
----------
dtype : str, `~numpy.dtype`, type
Input as an object that can be converted via :class:`numpy.dtype`.
Returns
-------
dtype_info_name : str
String name of ``dtype``
"""
dtype = np.dtype(dtype)
if dtype.names is not None:
info_names = ", ".join(dtype_info_name(dt[0]) for dt in dtype.fields.values())
return f"({info_names})"
if dtype.subdtype is not None:
dtype, shape = dtype.subdtype
else:
shape = ()
if dtype.kind in ("S", "U"):
type_name = "bytes" if dtype.kind == "S" else "str"
length = re.search(r"(\d+)", dtype.str).group(1)
out = type_name + length
else:
out = dtype.name
if shape:
out += f"[{','.join(str(n) for n in shape)}]"
return out
[docs]
def data_info_factory(names, funcs):
"""
Factory to create a function that can be used as an ``option``
for outputting data object summary information.
Examples
--------
>>> from astropy.utils.data_info import data_info_factory
>>> from astropy.table import Column
>>> c = Column([4., 3., 2., 1.])
>>> mystats = data_info_factory(names=['min', 'median', 'max'],
... funcs=[np.min, np.median, np.max])
>>> c.info(option=mystats)
min = 1
median = 2.5
max = 4
n_bad = 0
length = 4
Parameters
----------
names : list
List of information attribute names
funcs : list
List of functions that compute the corresponding information attribute
Returns
-------
func : function
Function that can be used as a data info option
"""
def func(dat):
outs = []
for name, func in zip(names, funcs):
try:
if isinstance(func, str):
out = getattr(dat, func)()
else:
out = func(dat)
except Exception:
outs.append("--")
else:
try:
outs.append(f"{out:g}")
except (TypeError, ValueError):
outs.append(str(out))
return OrderedDict(zip(names, outs))
return func
def _get_data_attribute(dat, attr=None):
"""
Get a data object attribute for the ``attributes`` info summary method.
"""
if attr == "class":
val = type(dat).__name__
elif attr == "dtype":
val = dtype_info_name(dat.info.dtype)
elif attr == "shape":
datshape = dat.shape[1:]
val = datshape if datshape else ""
else:
val = getattr(dat.info, attr)
if val is None:
val = ""
return str(val)
class InfoAttribute:
def __init__(self, attr, default=None):
self.attr = attr
self.default = default
def __get__(self, instance, owner_cls):
if instance is None:
return self
return instance._attrs.get(self.attr, self.default)
def __set__(self, instance, value):
if instance is None:
# This is an unbound descriptor on the class
raise ValueError("cannot set unbound descriptor")
instance._attrs[self.attr] = value
class ParentAttribute:
def __init__(self, attr):
self.attr = attr
def __get__(self, instance, owner_cls):
if instance is None:
return self
return getattr(instance._parent, self.attr)
def __set__(self, instance, value):
if instance is None:
# This is an unbound descriptor on the class
raise ValueError("cannot set unbound descriptor")
setattr(instance._parent, self.attr, value)
class DataInfoMeta(type):
def __new__(mcls, name, bases, dct):
# Ensure that we do not gain a __dict__, which would mean
# arbitrary attributes could be set.
dct.setdefault("__slots__", [])
return super().__new__(mcls, name, bases, dct)
def __init__(cls, name, bases, dct):
super().__init__(name, bases, dct)
# Define default getters/setters for attributes, if needed.
for attr in cls.attr_names:
if attr not in dct:
# If not defined explicitly for this class, did any of
# its superclasses define it, and, if so, was this an
# automatically defined look-up-on-parent attribute?
cls_attr = getattr(cls, attr, None)
if attr in cls.attrs_from_parent:
# If the attribute is supposed to be stored on the parent,
# and that is stated by this class yet it was not the case
# on the superclass, override it.
if "attrs_from_parent" in dct and not isinstance(
cls_attr, ParentAttribute
):
setattr(cls, attr, ParentAttribute(attr))
elif not cls_attr or isinstance(cls_attr, ParentAttribute):
# If the attribute is not meant to be stored on the parent,
# and if it was not defined already or was previously defined
# as an attribute on the parent, define a regular
# look-up-on-info attribute
setattr(
cls, attr, InfoAttribute(attr, cls._attr_defaults.get(attr))
)
[docs]
class DataInfo(metaclass=DataInfoMeta):
"""
Descriptor that data classes use to add an ``info`` attribute for storing
data attributes in a uniform and portable way. Note that it *must* be
called ``info`` so that the DataInfo() object can be stored in the
``instance`` using the ``info`` key. Because owner_cls.x is a descriptor,
Python doesn't use __dict__['x'] normally, and the descriptor can safely
store stuff there. Thanks to
https://nbviewer.jupyter.org/urls/gist.github.com/ChrisBeaumont/5758381/raw/descriptor_writeup.ipynb
for this trick that works for non-hashable classes.
Parameters
----------
bound : bool
If True this is a descriptor attribute in a class definition, else it
is a DataInfo() object that is bound to a data object instance. Default is False.
"""
_stats = ["mean", "std", "min", "max"]
attrs_from_parent = set()
attr_names = {"name", "unit", "dtype", "format", "description", "meta"}
_attr_defaults = {"dtype": np.dtype("O")}
_attrs_no_copy = set()
_info_summary_attrs = ("dtype", "shape", "unit", "format", "description", "class")
__slots__ = ["_parent_cls", "_parent_ref", "_attrs"]
# This specifies the list of object attributes which must be stored in
# order to re-create the object after serialization. This is independent
# of normal `info` attributes like name or description. Subclasses will
# generally either define this statically (QuantityInfo) or dynamically
# (SkyCoordInfo). These attributes may be scalars or arrays. If arrays
# that match the object length they will be serialized as an independent
# column.
_represent_as_dict_attrs = ()
# This specifies attributes which are to be provided to the class
# initializer as ordered args instead of keyword args. This is needed
# for Quantity subclasses where the keyword for data varies (e.g.
# between Quantity and Angle).
_construct_from_dict_args = ()
# This specifies the name of an attribute which is the "primary" data.
# Then when representing as columns
# (table.serialize._represent_mixin_as_column) the output for this
# attribute will be written with the just name of the mixin instead of the
# usual "<name>.<attr>".
_represent_as_dict_primary_data = None
def __init__(self, bound=False):
# If bound to a data object instance then create the dict of attributes
# which stores the info attribute values. Default of None for "unset"
# except for dtype where the default is object.
if bound:
self._attrs = {}
@property
def _parent(self):
try:
parent = self._parent_ref()
except AttributeError:
return None
if parent is None:
raise AttributeError(
"""\
failed to access "info" attribute on a temporary object.
It looks like you have done something like ``col[3:5].info`` or
``col.quantity.info``, i.e. you accessed ``info`` from a temporary slice
object that only exists momentarily. This has failed because the reference to
that temporary object is now lost. Instead force a permanent reference (e.g.
``c = col[3:5]`` followed by ``c.info``)."""
)
return parent
def __get__(self, instance, owner_cls):
if instance is None:
# This is an unbound descriptor on the class
self._parent_cls = owner_cls
return self
info = instance.__dict__.get("info")
if info is None:
info = instance.__dict__["info"] = self.__class__(bound=True)
# We set _parent_ref on every call, since if one makes copies of
# instances, 'info' will be copied as well, which will lose the
# reference.
info._parent_ref = weakref.ref(instance)
return info
def __set__(self, instance, value):
if instance is None:
# This is an unbound descriptor on the class
raise ValueError("cannot set unbound descriptor")
if isinstance(value, DataInfo):
info = instance.__dict__["info"] = self.__class__(bound=True)
attr_names = info.attr_names
if value.__class__ is self.__class__:
# For same class, attributes are guaranteed to be stored in
# _attrs, so speed matters up by not accessing defaults.
# Doing this before difference in for loop helps speed.
attr_names = attr_names & set(value._attrs) # NOT in-place!
else:
# For different classes, copy over the attributes in common.
attr_names = attr_names & (value.attr_names - value._attrs_no_copy)
for attr in attr_names - info.attrs_from_parent - info._attrs_no_copy:
info._attrs[attr] = deepcopy(getattr(value, attr))
else:
raise TypeError("info must be set with a DataInfo instance")
def __getstate__(self):
return self._attrs
def __setstate__(self, state):
self._attrs = state
def _represent_as_dict(self, attrs=None):
"""Get the values for the parent ``attrs`` and return as a dict. This
ignores any attributes that are None. In the context of serializing
the supported core astropy classes this conversion will succeed and
results in more succinct and less python-specific YAML.
By default, uses '_represent_as_dict_attrs'.
"""
return {
key: val
for key in (self._represent_as_dict_attrs if attrs is None else attrs)
if (val := getattr(self._parent, key, None)) is not None
}
def _construct_from_dict(self, map):
args = [map.pop(attr) for attr in self._construct_from_dict_args]
return self._parent_cls(*args, **map)
info_summary_attributes = staticmethod(
data_info_factory(
names=_info_summary_attrs,
funcs=[
partial(_get_data_attribute, attr=attr) for attr in _info_summary_attrs
],
)
)
# No nan* methods in numpy < 1.8
info_summary_stats = staticmethod(
data_info_factory(
names=_stats, funcs=[getattr(np, "nan" + stat) for stat in _stats]
)
)
[docs]
def __call__(self, option="attributes", out=""):
"""
Write summary information about data object to the ``out`` filehandle.
By default this prints to standard output via sys.stdout.
The ``option`` argument specifies what type of information
to include. This can be a string, a function, or a list of
strings or functions. Built-in options are:
- ``attributes``: data object attributes like ``dtype`` and ``format``
- ``stats``: basic statistics: min, mean, and max
If a function is specified then that function will be called with the
data object as its single argument. The function must return an
OrderedDict containing the information attributes.
If a list is provided then the information attributes will be
appended for each of the options, in order.
Examples
--------
>>> from astropy.table import Column
>>> c = Column([1, 2], unit='m', dtype='int32')
>>> c.info()
dtype = int32
unit = m
class = Column
n_bad = 0
length = 2
>>> c.info(['attributes', 'stats'])
dtype = int32
unit = m
class = Column
mean = 1.5
std = 0.5
min = 1
max = 2
n_bad = 0
length = 2
Parameters
----------
option : str, callable, list of (str or callable)
Info option, defaults to 'attributes'.
out : file-like, None
Output destination, defaults to sys.stdout. If None then the
OrderedDict with information attributes is returned
Returns
-------
info : `~collections.OrderedDict` or None
`~collections.OrderedDict` if out==None else None
"""
if out == "":
out = sys.stdout
dat = self._parent
info = OrderedDict()
name = dat.info.name
if name is not None:
info["name"] = name
options = option if isinstance(option, (list, tuple)) else [option]
for option_ in options:
if isinstance(option_, str):
if hasattr(self, "info_summary_" + option_):
option_ = getattr(self, "info_summary_" + option_)
else:
raise ValueError(
f"option={option_} is not an allowed information type"
)
with warnings.catch_warnings():
for ignore_kwargs in IGNORE_WARNINGS:
warnings.filterwarnings("ignore", **ignore_kwargs)
info.update(option_(dat))
if hasattr(dat, "mask"):
n_bad = np.count_nonzero(dat.mask)
else:
try:
n_bad = np.count_nonzero(np.isinf(dat) | np.isnan(dat))
except Exception:
n_bad = 0
info["n_bad"] = n_bad
try:
info["length"] = len(dat)
except (TypeError, IndexError):
pass
if out is None:
return info
for key, val in info.items():
if val != "":
out.write(f"{key} = {val}" + os.linesep)
def __repr__(self):
if self._parent is None:
return super().__repr__()
out = StringIO()
self.__call__(out=out)
return out.getvalue()
[docs]
class BaseColumnInfo(DataInfo):
"""Base info class for anything that can be a column in an astropy Table.
There are at least two classes that inherit from this:
ColumnInfo: for native astropy Column / MaskedColumn objects
MixinInfo: for mixin column objects
Note that this class is defined here so that mixins can use it
without importing the table package.
"""
attr_names = DataInfo.attr_names | {"parent_table", "indices"}
_attrs_no_copy = {"parent_table", "indices"}
# Context for serialization. This can be set temporarily via
# ``serialize_context_as(context)`` context manager to allow downstream
# code to understand the context in which a column is being serialized.
# Typical values are 'fits', 'hdf5', 'parquet', 'ecsv', 'yaml'. Objects
# like Time or SkyCoord will have different default serialization
# representations depending on context.
_serialize_context = None
__slots__ = ["_format_funcs", "_copy_indices"]
@property
def parent_table(self):
value = self._attrs.get("parent_table")
if callable(value):
value = value()
return value
@parent_table.setter
def parent_table(self, parent_table):
if parent_table is None:
self._attrs.pop("parent_table", None)
else:
parent_table = weakref.ref(parent_table)
self._attrs["parent_table"] = parent_table
def __init__(self, bound=False):
super().__init__(bound=bound)
# If bound to a data object instance then add a _format_funcs dict
# for caching functions for print formatting.
if bound:
self._format_funcs = {}
def __set__(self, instance, value):
# For Table columns do not set `info` when the instance is a scalar.
try:
if not instance.shape:
return
except AttributeError:
pass
super().__set__(instance, value)
[docs]
def iter_str_vals(self):
"""
This is a mixin-safe version of Column.iter_str_vals.
"""
col = self._parent
if self.parent_table is None:
from astropy.table.column import FORMATTER as formatter
else:
formatter = self.parent_table.formatter
_pformat_col_iter = formatter._pformat_col_iter
yield from _pformat_col_iter(col, -1, False, False, {})
@property
def indices(self):
# Implementation note: the auto-generation as an InfoAttribute cannot
# be used here, since on access, one should not just return the
# default (empty list is this case), but set _attrs['indices'] so that
# if the list is appended to, it is registered here.
return self._attrs.setdefault("indices", [])
@indices.setter
def indices(self, indices):
self._attrs["indices"] = indices
[docs]
def adjust_indices(self, index, value, col_len):
"""
Adjust info indices after column modification.
Parameters
----------
index : slice, int, list, or ndarray
Element(s) of column to modify. This parameter can
be a single row number, a list of row numbers, an
ndarray of row numbers, a boolean ndarray (a mask),
or a column slice.
value : int, list, or ndarray
New value(s) to insert
col_len : int
Length of the column
"""
if not self.indices:
return
if isinstance(index, slice):
# run through each key in slice
t = index.indices(col_len)
keys = list(range(*t))
elif isinstance(index, np.ndarray) and index.dtype.kind == "b":
# boolean mask
keys = np.where(index)[0]
else: # single int
keys = [index]
value = np.atleast_1d(value) # turn array(x) into array([x])
if value.size == 1:
# repeat single value
value = list(value) * len(keys)
for key, val in zip(keys, value):
for col_index in self.indices:
col_index.replace(key, self.name, val)
[docs]
def slice_indices(self, col_slice, item, col_len):
"""
Given a sliced object, modify its indices
to correctly represent the slice.
Parameters
----------
col_slice : `~astropy.table.Column` or mixin
Sliced object. If not a column, it must be a valid mixin, see
https://docs.astropy.org/en/stable/table/mixin_columns.html
item : slice, list, or ndarray
Slice used to create col_slice
col_len : int
Length of original object
"""
from astropy.table.sorted_array import SortedArray
if not getattr(self, "_copy_indices", True):
# Necessary because MaskedArray will perform a shallow copy
col_slice.info.indices = []
return col_slice
elif isinstance(item, slice):
col_slice.info.indices = [x[item] for x in self.indices]
elif self.indices:
if isinstance(item, np.ndarray) and item.dtype.kind == "b":
# boolean mask
item = np.where(item)[0]
# Empirical testing suggests that recreating a BST/RBT index is
# more effective than relabelling when less than ~60% of
# the total number of rows are involved, and is in general
# more effective for SortedArray.
small = len(item) <= 0.6 * col_len
col_slice.info.indices = []
for index in self.indices:
if small or isinstance(index, SortedArray):
new_index = index.get_slice(col_slice, item)
else:
new_index = deepcopy(index)
new_index.replace_rows(item)
col_slice.info.indices.append(new_index)
return col_slice
[docs]
@staticmethod
def merge_cols_attributes(cols, metadata_conflicts, name, attrs):
"""
Utility method to merge and validate the attributes ``attrs`` for the
input table columns ``cols``.
Note that ``dtype`` and ``shape`` attributes are handled specially.
These should not be passed in ``attrs`` but will always be in the
returned dict of merged attributes.
Parameters
----------
cols : list
List of input Table column objects
metadata_conflicts : str ('warn'|'error'|'silent')
How to handle metadata conflicts
name : str
Output column name
attrs : list
List of attribute names to be merged
Returns
-------
attrs : dict
Of merged attributes.
"""
from astropy.table.np_utils import TableMergeError
def warn_str_func(key, left, right):
out = (
f"In merged column '{name}' the '{key}' attribute does not match "
f"({left} != {right}). Using {right} for merged output"
)
return out
def getattrs(col):
return {
attr: getattr(col.info, attr)
for attr in attrs
if getattr(col.info, attr, None) is not None
}
out = getattrs(cols[0])
for col in cols[1:]:
out = metadata.merge(
out,
getattrs(col),
metadata_conflicts=metadata_conflicts,
warn_str_func=warn_str_func,
)
# Output dtype is the superset of all dtypes in in_cols
out["dtype"] = metadata.common_dtype(cols)
# Make sure all input shapes are the same
uniq_shapes = {col.shape[1:] for col in cols}
if len(uniq_shapes) != 1:
raise TableMergeError("columns have different shapes")
out["shape"] = uniq_shapes.pop()
# "Merged" output name is the supplied name
if name is not None:
out["name"] = name
return out
[docs]
def get_sortable_arrays(self):
"""
Return a list of arrays which can be lexically sorted to represent
the order of the parent column.
The base method raises NotImplementedError and must be overridden.
Returns
-------
arrays : list of ndarray
"""
raise NotImplementedError(f"column {self.name} is not sortable")
[docs]
class MixinInfo(BaseColumnInfo):
@property
def name(self):
return self._attrs.get("name")
@name.setter
def name(self, name):
# For mixin columns that live within a table, rename the column in the
# table when setting the name attribute. This mirrors the same
# functionality in the BaseColumn class.
if self.parent_table is not None:
new_name = None if name is None else str(name)
self.parent_table.columns._rename_column(self.name, new_name)
self._attrs["name"] = name
@property
def groups(self):
# This implementation for mixin columns essentially matches the Column
# property definition. `groups` is a read-only property here and
# depends on the parent table of the column having `groups`. This will
# allow aggregating mixins as long as they support those operations.
from astropy.table import groups
return self._attrs.setdefault("groups", groups.ColumnGroups(self._parent))
[docs]
class ParentDtypeInfo(MixinInfo):
"""Mixin that gets info.dtype from parent."""
attrs_from_parent = {"dtype"} # dtype and unit taken from parent