A Guide to NumPy/SciPy Documentation¶
When using Sphinx in combination with the
numpy conventions, you should use the numpydoc
extension so that your
docstrings will be handled correctly. For example, Sphinx will extract the
Parameters
section from your docstring and convert it into a field
list. Using numpydoc
will also avoid the reStructuredText errors produced
by plain Sphinx when it encounters numpy docstring conventions like
section headers (e.g. -------------
) that sphinx does not expect to
find in docstrings.
Some features described in this document require a recent version of
numpydoc
. For example, the Yields section was added in
numpydoc
0.6.
It is available from:
Note that for documentation within numpy, it is not necessary to do
import numpy as np
at the beginning of an example. However, some
sub-modules, such as fft
, are not imported by default, and you have to
include them explicitly:
import numpy.fft
after which you may use it:
np.fft.fft2(...)
For convenience the formatting standard is included below with an example
numpydoc docstring guide¶
This document describes the syntax and best practices for docstrings used with the numpydoc extension for Sphinx.
Note
For an accompanying example, see example.py.
Some features described in this document require a recent version of
numpydoc
. For example, the Yields section was added in
numpydoc
0.6.
Overview¶
- We mostly follow the standard Python style conventions as described here:
- Additional PEPs of interest regarding documentation of code:
- Use a code checker:
- pylint
- pyflakes
- pep8.py
- flake8
- vim-flake8 plugin for automatically checking syntax and style with flake8
Import conventions¶
The following import conventions are used throughout the NumPy source and documentation:
import numpy as np
import matplotlib as mpl
import matplotlib.pyplot as plt
Do not abbreviate scipy
. There is no motivating use case to
abbreviate it in the real world, so we avoid it in the documentation
to avoid confusion.
Docstring Standard¶
A documentation string (docstring) is a string that describes a module,
function, class, or method definition. The docstring is a special attribute
of the object (object.__doc__
) and, for consistency, is surrounded by
triple double quotes, i.e.:
"""This is the form of a docstring.
It can be spread over several lines.
"""
NumPy, SciPy, and the scikits follow a common convention for docstrings that provides for consistency, while also allowing our toolchain to produce well-formatted reference guides. This document describes the current community consensus for such a standard. If you have suggestions for improvements, post them on the numpy-discussion list.
Our docstring standard uses re-structured text (reST) syntax and is rendered using Sphinx (a pre-processor that understands the particular documentation style we are using). While a rich set of markup is available, we limit ourselves to a very basic subset, in order to provide docstrings that are easy to read on text-only terminals.
A guiding principle is that human readers of the text are given precedence over contorting docstrings so our tools produce nice output. Rather than sacrificing the readability of the docstrings, we have written pre-processors to assist Sphinx in its task.
The length of docstring lines should be kept to 75 characters to facilitate reading the docstrings in text terminals.
Sections¶
The docstring consists of a number of sections separated by headings (except for the deprecation warning). Each heading should be underlined in hyphens, and the section ordering should be consistent with the description below.
The sections of a function’s docstring are:
Short summary
A one-line summary that does not use variable names or the function name, e.g.
def add(a, b): """The sum of two numbers. """
The function signature is normally found by introspection and displayed by the help function. For some functions (notably those written in C) the signature is not available, so we have to specify it as the first line of the docstring:
""" add(a, b) The sum of two numbers. """
Deprecation warning
A section (use if applicable) to warn users that the object is deprecated. Section contents should include:
- In what NumPy version the object was deprecated, and when it will be removed.
- Reason for deprecation if this is useful information (e.g., object is superseded, duplicates functionality found elsewhere, etc.).
- New recommended way of obtaining the same functionality.
This section should use the
deprecated
Sphinx directive instead of an underlined section header... deprecated:: 1.6.0 `ndobj_old` will be removed in NumPy 2.0.0, it is replaced by `ndobj_new` because the latter works also with array subclasses.
Extended Summary
A few sentences giving an extended description. This section should be used to clarify functionality, not to discuss implementation detail or background theory, which should rather be explored in the Notes section below. You may refer to the parameters and the function name, but parameter descriptions still belong in the Parameters section.
Parameters
Description of the function arguments, keywords and their respective types.
Parameters ---------- x : type Description of parameter `x`. y Description of parameter `y` (with type not specified)
Enclose variables in single backticks. The colon must be preceded by a space, or omitted if the type is absent.
For the parameter types, be as precise as possible. Below are a few examples of parameters and their types.
Parameters ---------- filename : str copy : bool dtype : data-type iterable : iterable object shape : int or tuple of int files : list of str
If it is not necessary to specify a keyword argument, use
optional
:x : int, optional
Optional keyword parameters have default values, which are displayed as part of the function signature. They can also be detailed in the description:
Description of parameter `x` (the default is -1, which implies summation over all axes).
When a parameter can only assume one of a fixed set of values, those values can be listed in braces, with the default appearing first:
order : {'C', 'F', 'A'} Description of `order`.
When two or more input parameters have exactly the same type, shape and description, they can be combined:
x1, x2 : array_like Input arrays, description of `x1`, `x2`.
Returns
Explanation of the returned values and their types. Similar to the Parameters section, except the name of each return value is optional. The type of each return value is always required:
Returns ------- int Description of anonymous integer return value.
If both the name and type are specified, the Returns section takes the same form as the Parameters section:
Returns ------- err_code : int Non-zero value indicates error code, or zero on success. err_msg : str or None Human readable error message, or None on success.
Yields
Explanation of the yielded values and their types. This is relevant to generators only. Similar to the Returns section in that the name of each value is optional, but the type of each value is always required:
Yields ------ int Description of the anonymous integer return value.
If both the name and type are specified, the Yields section takes the same form as the Returns section:
Yields ------ err_code : int Non-zero value indicates error code, or zero on success. err_msg : str or None Human readable error message, or None on success.
Support for the Yields section was added in numpydoc version 0.6.
Other Parameters
An optional section used to describe infrequently used parameters. It should only be used if a function has a large number of keyword parameters, to prevent cluttering the Parameters section.
Raises
An optional section detailing which errors get raised and under what conditions:
Raises ------ LinAlgException If the matrix is not numerically invertible.
This section should be used judiciously, i.e., only for errors that are non-obvious or have a large chance of getting raised.
Warns
An optional section detailing which warnings get raised and under what conditions, formatted similarly to Raises.
Warnings
An optional section with cautions to the user in free text/reST.
- See Also
An optional section used to refer to related code. This section can be very useful, but should be used judiciously. The goal is to direct users to other functions they may not be aware of, or have easy means of discovering (by looking at the module docstring, for example). Routines whose docstrings further explain parameters used by this function are good candidates.
As an example, for
numpy.mean
we would have:See Also -------- average : Weighted averageWhen referring to functions in the same sub-module, no prefix is needed, and the tree is searched upwards for a match.
Prefix functions from other sub-modules appropriately. E.g., whilst documenting the
random
module, refer to a function infft
byfft.fft2 : 2-D fast discrete Fourier transformWhen referring to an entirely different module:
scipy.random.norm : Random variates, PDFs, etc.Functions may be listed without descriptions, and this is preferable if the functionality is clear from the function name:
See Also -------- func_a : Function a with its description. func_b, func_c_, func_d func_e
Notes
An optional section that provides additional information about the code, possibly including a discussion of the algorithm. This section may include mathematical equations, written in LaTeX format:
The FFT is a fast implementation of the discrete Fourier transform: .. math:: X(e^{j\omega } ) = x(n)e^{ - j\omega n}
Equations can also be typeset underneath the math directive:
The discrete-time Fourier time-convolution property states that .. math:: x(n) * y(n) \Leftrightarrow X(e^{j\omega } )Y(e^{j\omega } )\\ another equation here
Math can furthermore be used inline, i.e.
The value of :math:`\omega` is larger than 5.
Variable names are displayed in typewriter font, obtained by using
\mathtt{var}
:We square the input parameter `alpha` to obtain :math:`\mathtt{alpha}^2`.
Note that LaTeX is not particularly easy to read, so use equations sparingly.
Images are allowed, but should not be central to the explanation; users viewing the docstring as text must be able to comprehend its meaning without resorting to an image viewer. These additional illustrations are included using:
.. image:: filename
where filename is a path relative to the reference guide source directory.
References
References cited in the notes section may be listed here, e.g. if you cited the article below using the text
[1]_
, include it as in the list as follows:.. [1] O. McNoleg, "The integration of GIS, remote sensing, expert systems and adaptive co-kriging for environmental habitat modelling of the Highland Haggis using object-oriented, fuzzy-logic and neural-network techniques," Computers & Geosciences, vol. 22, pp. 585-588, 1996.
which renders as [1]:
[1] O. McNoleg, “The integration of GIS, remote sensing, expert systems and adaptive co-kriging for environmental habitat modelling of the Highland Haggis using object-oriented, fuzzy-logic and neural-network techniques,” Computers & Geosciences, vol. 22, pp. 585-588, 1996.
Referencing sources of a temporary nature, like web pages, is discouraged. References are meant to augment the docstring, but should not be required to understand it. References are numbered, starting from one, in the order in which they are cited.
Warning
References will break tables
Where references like [1] appear in a tables within a numpydoc docstring, the table markup will be broken by numpydoc processing. See numpydoc issue #130
Examples
An optional section for examples, using the doctest format. This section is meant to illustrate usage, not to provide a testing framework – for that, use the
tests/
directory. While optional, this section is very strongly encouraged.When multiple examples are provided, they should be separated by blank lines. Comments explaining the examples should have blank lines both above and below them:
>>> np.add(1, 2) 3 Comment explaining the second example >>> np.add([1, 2], [3, 4]) array([4, 6])
For tests with a result that is random or platform-dependent, mark the output as such:
>>> import numpy.random >>> np.random.rand(2) array([ 0.35773152, 0.38568979]) #random
You can run examples as doctests using:
>>> np.test(doctests=True) >>> np.linalg.test(doctests=True) # for a single module
In IPython it is also possible to run individual examples simply by copy-pasting them in doctest mode:
In [1]: %doctest_mode Exception reporting mode: Plain Doctest mode is: ON >>> %paste import numpy.random np.random.rand(2) ## -- End pasted text -- array([ 0.8519522 , 0.15492887])
It is not necessary to use the doctest markup
<BLANKLINE>
to indicate empty lines in the output. Note that the option to run the examples throughnumpy.test
is provided for checking if the examples work, not for making the examples part of the testing framework.The examples may assume that
import numpy as np
is executed before the example code in numpy. Additional examples may make use of matplotlib for plotting, but should import it explicitly, e.g.,import matplotlib.pyplot as plt
. All other imports, including the demonstrated function, must be explicit.When matplotlib is imported in the example, the Example code will be wrapped in matplotlib’s Sphinx `plot directive <http://matplotlib.org/sampledoc/extensions.html>`_. When matplotlib is not explicitly imported, plot:: can be used directly if
matplotlib.sphinxext.plot_directive
is loaded as a Sphinx extension inconf.py
.
Documenting classes¶
Class docstring¶
Use the same sections as outlined above (all except Returns
are
applicable). The constructor (__init__
) should also be documented
here, the Parameters section of the docstring details the constructors
parameters.
An Attributes section, located below the Parameters section, may be used to describe non-method attributes of the class:
Attributes
----------
x : float
The X coordinate.
y : float
The Y coordinate.
Attributes that are properties and have their own docstrings can be simply listed by name:
Attributes
----------
real
imag
x : float
The X coordinate
y : float
The Y coordinate
In general, it is not necessary to list class methods. Those that are not part of the public API have names that start with an underscore. In some cases, however, a class may have a great many methods, of which only a few are relevant (e.g., subclasses of ndarray). Then, it becomes useful to have an additional Methods section:
class Photo(ndarray):
"""
Array with associated photographic information.
...
Attributes
----------
exposure : float
Exposure in seconds.
Methods
-------
colorspace(c='rgb')
Represent the photo in the given colorspace.
gamma(n=1.0)
Change the photo's gamma exposure.
"""
If it is necessary to explain a private method (use with care!), it can be referred to in the Extended Summary or the Notes section. Do not list private methods in the methods section.
Note that self is not listed as the first parameter of methods.
Method docstrings¶
Document these as you would any other function. Do not include
self
in the list of parameters. If a method has an equivalent function
(which is the case for many ndarray methods for example), the function
docstring should contain the detailed documentation, and the method docstring
should refer to it. Only put brief summary and See Also sections in the
method docstring. The method should use a Returns or Yields section,
as appropriate.
Documenting class instances¶
Instances of classes that are part of the NumPy API (for example np.r_ np,c_, np.index_exp, etc.) may require some care. To give these instances a useful docstring, we do the following:
- Single instance: If only a single instance of a class is exposed, document the class. Examples can use the instance name.
- Multiple instances: If multiple instances are exposed, docstrings
for each instance are written and assigned to the instances’
__doc__
attributes at run time. The class is documented as usual, and the exposed instances can be mentioned in the Notes and See Also sections.
Documenting generators¶
Generators should be documented just as functions are documented. The only difference is that one should use the Yields section instead of the Returns section. Support for the Yields section was added in numpydoc version 0.6.
Documenting constants¶
Use the same sections as outlined for functions where applicable:
1. summary
2. extended summary (optional)
3. see also (optional)
4. references (optional)
5. examples (optional)
Docstrings for constants will not be visible in text terminals (constants are of immutable type, so docstrings can not be assigned to them like for for class instances), but will appear in the documentation built with Sphinx.
Documenting modules¶
Each module should have a docstring with at least a summary line. Other sections are optional, and should be used in the same order as for documenting functions when they are appropriate:
1. summary
2. extended summary
3. routine listings
4. see also
5. notes
6. references
7. examples
Routine listings are encouraged, especially for large modules, for which it is
hard to get a good overview of all functionality provided by looking at the
source file(s) or the __all__
dict.
Note that license and author info, while often included in source files, do not belong in docstrings.
Other points to keep in mind¶
Equations : as discussed in the Notes section above, LaTeX formatting should be kept to a minimum. Often it’s possible to show equations as Python code or pseudo-code instead, which is much more readable in a terminal. For inline display use double backticks (like
y = np.sin(x)
). For display with blank lines above and below, use a double colon and indent the code, like:end of previous sentence:: y = np.sin(x)
Notes and Warnings : If there are points in the docstring that deserve special emphasis, the reST directives for a note or warning can be used in the vicinity of the context of the warning (inside a section). Syntax:
.. warning:: Warning text. .. note:: Note text.
Use these sparingly, as they do not look very good in text terminals and are not often necessary. One situation in which a warning can be useful is for marking a known bug that is not yet fixed.
array_like : For functions that take arguments which can have not only a type ndarray, but also types that can be converted to an ndarray (i.e. scalar types, sequence types), those arguments can be documented with type array_like.
Links : If you need to include hyperlinks in your docstring, note that some docstring sections are not parsed as standard reST, and in these sections, numpydoc may become confused by hyperlink targets such as:
.. _Example: http://www.example.com
If the Sphinx build issues a warning of the form
WARNING: Unknown target name: "example"
, then that is what is happening. To avoid this problem, use the inline hyperlink form:`Example <http://www.example.com>`_
Common reST concepts¶
For paragraphs, indentation is significant and indicates indentation in the output. New paragraphs are marked with a blank line.
Use *italics*
, **bold**
and ``monospace``
if needed in any
explanations
(but not for variable names and doctest code or multi-line code).
Variable, module, function, and class names should be written between
single back-ticks (`numpy`
).
A more extensive example of reST markup can be found in this example document; the quick reference is useful while editing.
Line spacing and indentation are significant and should be carefully followed.
Conclusion¶
This document itself was written in ReStructuredText. An example of the format shown here is available.
Example Source¶
"""This is the docstring for the example.py module. Modules names should
have short, all-lowercase names. The module name may have underscores if
this improves readability.
Every module should have a docstring at the very top of the file. The
module's docstring may extend over multiple lines. If your docstring does
extend over multiple lines, the closing three quotation marks must be on
a line by itself, preferably preceded by a blank line.
"""
from __future__ import division, absolute_import, print_function
import os # standard library imports first
# Do NOT import using *, e.g. from numpy import *
#
# Import the module using
#
# import numpy
#
# instead or import individual functions as needed, e.g
#
# from numpy import array, zeros
#
# If you prefer the use of abbreviated module names, we suggest the
# convention used by NumPy itself::
import numpy as np
import matplotlib as mpl
import matplotlib.pyplot as plt
# These abbreviated names are not to be used in docstrings; users must
# be able to paste and execute docstrings after importing only the
# numpy module itself, unabbreviated.
def foo(var1, var2, long_var_name='hi'):
r"""A one-line summary that does not use variable names or the
function name.
Several sentences providing an extended description. Refer to
variables using back-ticks, e.g. `var`.
Parameters
----------
var1 : array_like
Array_like means all those objects -- lists, nested lists, etc. --
that can be converted to an array. We can also refer to
variables like `var1`.
var2 : int
The type above can either refer to an actual Python type
(e.g. ``int``), or describe the type of the variable in more
detail, e.g. ``(N,) ndarray`` or ``array_like``.
long_var_name : {'hi', 'ho'}, optional
Choices in brackets, default first when optional.
Returns
-------
type
Explanation of anonymous return value of type ``type``.
describe : type
Explanation of return value named `describe`.
out : type
Explanation of `out`.
type_without_description
Other Parameters
----------------
only_seldom_used_keywords : type
Explanation
common_parameters_listed_above : type
Explanation
Raises
------
BadException
Because you shouldn't have done that.
See Also
--------
otherfunc : relationship (optional)
newfunc : Relationship (optional), which could be fairly long, in which
case the line wraps here.
thirdfunc, fourthfunc, fifthfunc
Notes
-----
Notes about the implementation algorithm (if needed).
This can have multiple paragraphs.
You may include some math:
.. math:: X(e^{j\omega } ) = x(n)e^{ - j\omega n}
And even use a greek symbol like :math:`omega` inline.
References
----------
Cite the relevant literature, e.g. [1]_. You may also cite these
references in the notes section above.
.. [1] O. McNoleg, "The integration of GIS, remote sensing,
expert systems and adaptive co-kriging for environmental habitat
modelling of the Highland Haggis using object-oriented, fuzzy-logic
and neural-network techniques," Computers & Geosciences, vol. 22,
pp. 585-588, 1996.
Examples
--------
These are written in doctest format, and should illustrate how to
use the function.
>>> a = [1, 2, 3]
>>> print [x + 3 for x in a]
[4, 5, 6]
>>> print "a\n\nb"
a
b
"""
pass
Example Rendered¶
This is the docstring for the example.py module. Modules names should have short, all-lowercase names. The module name may have underscores if this improves readability.
Every module should have a docstring at the very top of the file. The module’s docstring may extend over multiple lines. If your docstring does extend over multiple lines, the closing three quotation marks must be on a line by itself, preferably preceded by a blank line.
-
doc.example.
foo
(var1, var2, long_var_name='hi')[source]¶ A one-line summary that does not use variable names or the function name.
Several sentences providing an extended description. Refer to variables using back-ticks, e.g. var.
Parameters: - var1 : array_like
Array_like means all those objects – lists, nested lists, etc. – that can be converted to an array. We can also refer to variables like var1.
- var2 : int
The type above can either refer to an actual Python type (e.g.
int
), or describe the type of the variable in more detail, e.g.(N,) ndarray
orarray_like
.- long_var_name : {‘hi’, ‘ho’}, optional
Choices in brackets, default first when optional.
Returns: - type
Explanation of anonymous return value of type
type
.- describe : type
Explanation of return value named describe.
- out : type
Explanation of out.
- type_without_description
Other Parameters: - only_seldom_used_keywords : type
Explanation
- common_parameters_listed_above : type
Explanation
Raises: - BadException
Because you shouldn’t have done that.
See also
otherfunc
- relationship (optional)
newfunc
- Relationship (optional), which could be fairly long, in which case the line wraps here.
thirdfunc
,fourthfunc
,fifthfunc
Notes
Notes about the implementation algorithm (if needed).
This can have multiple paragraphs.
You may include some math:
X(e^{j\omega } ) = x(n)e^{ - j\omega n}
And even use a greek symbol like omega inline.
References
Cite the relevant literature, e.g. [1]. You may also cite these references in the notes section above.
[1] (1, 2) O. McNoleg, “The integration of GIS, remote sensing, expert systems and adaptive co-kriging for environmental habitat modelling of the Highland Haggis using object-oriented, fuzzy-logic and neural-network techniques,” Computers & Geosciences, vol. 22, pp. 585-588, 1996. Examples
These are written in doctest format, and should illustrate how to use the function.
>>> a = [1, 2, 3] >>> print [x + 3 for x in a] [4, 5, 6] >>> print "a\n\nb" a b