10000 DOC: Forward port 1.7.0 release notes · numpy/numpy@143fb18 · GitHub
[go: up one dir, main page]
Skip to content

Commit 143fb18

Browse files
charrischarris
committed
DOC: Forward port 1.7.0 release notes
1 parent 7eec4e3 commit 143fb18
< 8000 div class="d-flex flex-column gap-2" id="diff_file_tree" tabindex="-1">

File tree

1 file changed

+279
-0
lines changed

1 file changed

+279
-0
lines changed

‎doc/release/1.7.0-notes.rst

Lines changed: 279 additions & 0 deletions
< 8000 td data-grid-cell-id="diff-fe5af5965c7aee100dcc9f1c94b5bd559da54afa79d1d751f3b3e4d02ba58d08-empty-42-0" data-selected="false" role="gridcell" style="background-color:var(--diffBlob-additionNum-bgColor, var(--diffBlob-addition-bgColor-num));text-align:center" tabindex="-1" valign="top" class="focusable-grid-cell diff-line-number position-relative left-side">
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,279 @@
1+
=========================
2+
NumPy 1.7.0 Release Notes
3+
=========================
4+
5+
This release includes several new features as well as numerous bug fixes and
6+
refactorings. It supports Python 2.4 - 2.7 and 3.1 - 3.2.
7+
8+
Highlights
9+
==========
10+
11+
* ``where=`` parameter to ufuncs (allows the use of boolean arrays to choose
12+
where a computation should be done)
13+
* ``vectorize`` improvements (added 'excluded' and 'cache' keyword, general
14+
cleanup and bug fixes)
15+
* ``numpy.random.choice`` (random sample generating function)
16+
17+
18+
Compatibility notes
19+
===================
20+
21+
In a future version of numpy, the functions np.diag, np.diagonal, and
22+
the diagonal method of ndarrays will return a view onto the original
23+
array, instead of producing a copy as they do now. This makes a
24+
difference if you write to the array returned by any of these
25+
functions. To facilitate this transition, numpy 1.7 produces a
26+
DeprecationWarning if it detects that you may be attempting to write
27+
to such an array. See the documentation for np.diagonal for details.
28+
29+
The default casting rule for UFunc out= parameters has been changed from
30+
'unsafe' to 'same_kind'. Most usages which violate the 'same_kind'
31+
rule are likely bugs, so this change may expose previously undetected
32+
errors in projects that depend on NumPy.
33+
34+
Full-array boolean indexing has been optimized to use a different,
35+
optimized code path. This code path should produce the same results,
36+
but any feedback about changes to your code would be appreciated.
37+
38+
Attempting to write to a read-only array (one with
39+
``arr.flags.writeable`` set to ``False``) used to raise either a
40+
RuntimeError, ValueError, or TypeError inconsistently, depending on
41+
which code path was taken. It now consistently raises a ValueError.
42+
43+
The <ufunc>.reduce functions evaluate some reductions in a different
44+
order than in previous versions of NumPy, generally providing higher
45+
performance. Because of the nature of floating-point arithmetic, this
46+
may subtly change some results, just as linking NumPy to a different
47+
BLAS implementations such as MKL can.
48+
49+
If upgrading from 1.5, then generally in 1.6 and 1.7 there have been
50+
substantial code added and some code paths altered, particularly in
51+
the areas of type resolution and buffered iteration over universal
52+
functions. This might have an impact on your code particularly if
53+
you relied on accidental behavior in the past.
54+
55+
New features
56+
============
57+
58+
Mask-based NA missing values
59+
----------------------------
60+
61+
Preliminary support for NA missing values similar to those in R has
62+
been implemented. This was done by adding optional NA masks to the core
63+
array object.
64+
65+
.. note:: The NA API is *experimental*, and may undergo changes in future
66+
versions of NumPy. The current implementation based on masks will likely be
67+
supplemented by a second one based on bit-patterns, and it is possible that
68+
a difference will be made between missing and ignored data.
69+
70+
While a significant amount of the NumPy functionality has been extended to
71+
support NA masks, not everything is yet supported. Here is an (incomplete)
72+
list of things that do and do not work with NA values:
73+
74+
What works with NA:
75+
* Basic indexing and slicing, as well as full boolean mask indexing.
76+
* All element-wise ufuncs.
77+
* All UFunc.reduce methods, with a new skipna parameter.
78+
* np.all and np.any satisfy the rules NA | True == True and
79+
NA & False == False
80+
* The nditer object.
81+
* Array methods:
82+
+ ndarray.clip, ndarray.min, ndarray.max, ndarray.sum, ndarray.prod,
83+
ndarray.conjugate, ndarray.diagonal, ndarray.flatten
84+
+ numpy.concatenate, numpy.column_stack, numpy.hstack,
85+
numpy.vstack, numpy.dstack, numpy.squeeze, numpy.mean, numpy.std,
86+
numpy.var
87+
88+
What doesn't work with NA:
89+
* Fancy indexing, such as with lists and partial boolean masks.
90+
* ndarray.flat and any other methods that use the old iterator
91+
mechanism instead of the newer nditer.
92+
* Struct dtypes, which will have corresponding struct masks with
93+
one mask value per primitive field of the struct dtype.
94+
* UFunc.accumulate, UFunc.reduceat.
95+
* Ufunc calls with both NA masks and a where= mask at the same time.
96+
* File I/O has not been extended to support NA-masks.
97+
* np.logical_and and np.logical_or don't satisfy the
98+
rules NA | True == True and NA & False == False yet.
99+
* Array methods:
100+
+ ndarray.argmax, ndarray.argmin,
101+
+ numpy.repeat, numpy.delete (relies on fancy indexing),
102+
numpy.append, numpy.insert (relies on fancy indexing),
103+
numpy.where,
104+
105+
Differences with R:
106+
* R's parameter rm.na=T is spelled skipna=True in NumPy.
107+
* np.isna(nan) is False, but R's is.na(nan) is TRUE. This is because
108+
NumPy's NA is treated independently of the underlying data type.
109+
* Boolean indexing, where the result is compressed to just
110+
the elements with true in the mask, raises if the boolean mask
111+
has an NA value in it. This is because that value could be either
112+
True or False, meaning the count of the output array is actually
113+
NA. R treats this case in a manner inconsistent with the NA model,
114+
returning NA values in the spots where the boolean index has NA.
115+
This may have a practical advantage in spite of violating the
116+
NA theoretical model, so NumPy could adopt the behavior if necessary
117+
118+
For more information, see `doc/neps/missing-data.rst <https://github.com/numpy/numpy/blob/maintenance/1.7.x/doc/neps/missing-data.rst>`_.
119+
120+
Reduction UFuncs Generalize axis= Parameter
121+
-------------------------------------------
122+
123+
Any ufunc.reduce function call, as well as other reductions like
124+
sum, prod, any, all, max and min support the ability to choose
125+
a subset of the axes to reduce over. Previously, one could say
126+
axis=None to mean all the axes or axis=# to pick a single axis.
127+
Now, one can also say axis=(#,#) to pick a list of axes for reduction.
128+
129+
Reduction UFuncs New keepdims= Parameter
130+
----------------------------------------
131+
132+
There is a new keepdims= parameter, which if set to True, doesn't
133+
throw away the reduction axes but instead sets them to have size one.
134+
When this option is set, the reduction result will broadcast correctly
135+
to the original operand which was reduced.
136+
137+
Datetime support
138+
----------------
139+
140+
.. note:: The datetime API is *experimental* in 1.7.0, and may undergo changes
141+
in future versions of NumPy.
142+
143+
There have been a lot of fixes and enhancements to datetime64 compared
144+
to NumPy 1.6:
145+
146+
* the parser is quite strict about only accepting ISO 8601 dates, with a few
147+
convenience extensions
148+
* converts between units correctly
149+
* datetime arithmetic works correctly
150+
* business day functionality (allows the datetime to be used in contexts where
151+
only certain days of the week are valid)
152+
153+
The notes in `doc/source/reference/arrays.datetime.rst <https://github.com/numpy/numpy/blob/maintenance/1.7.x/doc/source/reference/arrays.datetime.rst>`_
154+
(also available in the online docs at `arrays.datetime.html
155+
<http://docs.scipy.org/doc/numpy/reference/arrays.datetime.html>`_) should be
156+
consulted for more details.
157+
158+
Custom formatter for printing arrays
159+
------------------------------------
160+
161+
See the new ``formatter`` parameter of the ``numpy.set_printoptions``
162+
function.
163+
164+
New function numpy.random.choice
165+
---------------------------------
166+
167+
A generic sampling function has been added which will generate samples from
168+
a given array-like. The samples can be with or without replacement, and
169+
with uniform or given non-uniform probabilities.
170+
171+
New function isclose
172+
--------------------
173+
174+
Returns a boolean array where two arrays are element-wise equal within a
175+
tolerance. Both relative and absolute tolerance can be specified. The
176+
function is NA aware.
177+
178+
Preliminary multi-dimensional support in the polynomial package
179+
---------------------------------------------------------------
180+
181+
Axis keywords have been added to the integration and differentiation
182+
functions and a tensor keyword was added to the evaluation functions.
183+
These additions allow multi-dimensional coefficient arrays to be used in
184+
those functions. New functions for evaluating 2-D and 3-D coefficient
185+
arrays on grids or sets of points were added together with 2-D and 3-D
186+
pseudo-Vandermonde matrices that can be used for fitting.
187+
188+
Support for mask-based NA values in the polynomial package fits
189+
---------------------------------------------------------------
190+
191+
The fitting functions recognize and remove masked data from the fit.
192+
193+
Ability to pad rank-n arrays
194+
----------------------------
195+
196+
A pad module containing functions for padding n-dimensional arrays has
197+
been added. The various private padding functions are exposed as options to
198+
a public 'pad' function. Example::
199+
200+
pad(a, 5, mode='mean')
201+
202+
Current modes are ``constant``, ``edge``, ``linear_ramp``, ``maximum``,
203+
``mean``, ``median``, ``minimum``, ``reflect``, ``symmetric``, ``wrap``, and
204+
``<function>``.
205+
206+
207+
New argument to searchsorted
208+
----------------------------
209+
210+
The function searchsorted now accepts a 'sorter' argument that is a
211+
permuation array that sorts the array to search.
212+
213+
C API
214+
-----
215+
216+
New function ``PyArray_RequireWriteable`` provides a consistent
217+
interface for checking array writeability -- any C code which works
218+
with arrays whose WRITEABLE flag is not known to be True a priori,
219+
should make sure to call this function before writing.
220+
221+
NumPy C Style Guide added (``doc/C_STYLE_GUIDE.rst.txt``).
222+
223+
Changes
224+
=======
225+
226+
General
227+
-------
228+
229+
The function np.concatenate tries to match the layout of its input
230+
arrays. Previously, the layout did not follow any particular reason,
231+
and depended in an undesirable way on the particular axis chosen for
232+
concatenation. A bug was also fixed which silently allowed out of bounds
233+
axis arguments.
234+
235+
The ufuncs logical_or, logical_and, and logical_not now follow Python's
236+
behavior with object arrays, instead of trying to call methods on the
237+
objects. For example the expression (3 and 'test') produces the string
238+
'test', and now np.logical_and(np.array(3, 'O'), np.array('test', 'O'))
239+
produces 'test' as well.
240+
241+
C-API
242+
-----
243+
244+
The following macros now require trailing semicolons::
245+
246+
NPY_BEGIN_THREADS_DEF
247+
NPY_BEGIN_THREADS
248+
NPY_ALLOW_C_API
249+
NPY_ALLOW_C_API_DEF
250+
NPY_DISABLE_C_API
251+
252+
253+
Deprecations
254+
============
255+
256+
General
257+
-------
258+
259+
Specifying a custom string formatter with a `_format` array attribute is
260+
deprecated. The new `formatter` keyword in ``numpy.set_printoptions`` or
261+
``numpy.array2string`` can be used instead.
262+
263+
The deprecated imports in the polynomial package have been removed.
264+
265+
C-API
266+
-----
267+
268+
Direct access to the fields of PyArrayObject* has been deprecated. Direct
269+
access has been recommended against for many releases. Expect similar
270+
deprecations for PyArray_Descr* and other core objects in the future as
271+
preparation for NumPy 2.0.
272+
273+
The macros in old_defines.h are deprecated and will be removed in the next
274+
minor release (>= 1.8). The sed script tools/replace_old_macros.sed can
275+
be used to replace these macros with the newer versions.
276+
277+
You can test your code against the deprecated C API by #defining
278+
NPY_NO_DEPRECATED_API to the target version number, for example
279+
NPY_1_7_API_VERSION, before including any NumPy headers.

0 commit comments

Comments
 (0)
0