Mailman 3 bug in ndarray.resize? - NumPy-Discussion

newer
Warnings in numpy.ma.test()

bug in ndarray.resize?

older
lists of zeros and ones

Alan G Isaac

17 Mar 2010 17 Mar '10

7:08 a.m.

Is the zero-fill intentional? If so, it is documented? (NumPy 1.3) Alan Isaac

...

...
...
a = np.arange(5) b = a.copy() c = np.resize(a, (5,2)) b.resize((5,2)) c # as expected array([[0, 1], [2, 3], [4, 0], [1, 2], [3, 4]]) b # surprise! array([[0, 1], [2, 3], [4, 0], [0, 0], [0, 0]])

Show replies by date

josef.pktd＠gmail.com

17 Mar 17 Mar

7:16 a.m.

On Wed, Mar 17, 2010 at 10:08 AM, Alan G Isaac wrote:

...

Is the zero-fill intentional? If so, it is documented? (NumPy 1.3)

Alan Isaac

...
...
...
a = np.arange(5) b = a.copy() c = np.resize(a, (5,2)) b.resize((5,2)) c # as expected array([[0, 1], [2, 3], [4, 0], [1, 2], [3, 4]]) b # surprise! array([[0, 1], [2, 3], [4, 0], [0, 0], [0, 0]])

It is documented as in your example numpy.resize(a, new_shape) Return a new array with the specified shape. If the new array is larger than the original array, then the new array is filled with repeated copied of a. Note that this behavior is different from a.resize(new_shape) which fills with zeros instead of repeated copies of a. Josef

...

_______________________________________________ NumPy-Discussion mailing list NumPy-Discussion@scipy.org http://mail.scipy.org/mailman/listinfo/numpy-discussion

Alan G Isaac

7:42 a.m.

On 3/17/2010 10:16 AM, josef.pktd@gmail.com wrote:

...

numpy.resize(a, new_shape) Return a new array with the specified shape.

If the new array is larger than the original array, then the new array is filled with repeated copied of a. Note that this behavior is different from a.resize(new_shape) which fills with zeros instead of repeated copies of a.

Yes indeed. Sorry, I must have scrolled the help without realizing it, and this part was at the top. So my follow up: why is this desirable/necessary? (I find it surprising.) Alan

Charles R Harris

2:56 p.m.

On Wed, Mar 17, 2010 at 8:42 AM, Alan G Isaac wrote:

...

On 3/17/2010 10:16 AM, josef.pktd@gmail.com wrote:

...
numpy.resize(a, new_shape) Return a new array with the specified shape.

If the new array is larger than the original array, then the new array is filled with repeated copied of a. Note that this behavior is different from a.resize(new_shape) which fills with zeros instead of repeated copies of a.

Yes indeed. Sorry, I must have scrolled the help without realizing it, and this part was at the top.

So my follow up: why is this desirable/necessary? (I find it surprising.)

IIRC, it behaved that way in Numeric. Chuck

Francesc Alted

18 Mar 18 Mar

1:55 a.m.

A Wednesday 17 March 2010 22:56:20 Charles R Harris escrigué:

...

On Wed, Mar 17, 2010 at 8:42 AM, Alan G Isaac wrote:

...
On 3/17/2010 10:16 AM, josef.pktd@gmail.com wrote:

...
numpy.resize(a, new_shape) Return a new array with the specified shape.

If the new array is larger than the original array, then the new array is filled with repeated copied of a. Note that this behavior is different from a.resize(new_shape) which fills with zeros instead of repeated copies of a.

Yes indeed. Sorry, I must have scrolled the help without realizing it, and this part was at the top.

So my follow up: why is this desirable/necessary? (I find it surprising.)

IIRC, it behaved that way in Numeric.

This does not mean that this behaviour is desirable. I find it inconsistent and misleading so +1 for fixing it. -- Francesc Alted

Sebastian Haase

1:56 a.m.

On Wed, Mar 17, 2010 at 10:56 PM, Charles R Harris wrote:

...

On Wed, Mar 17, 2010 at 8:42 AM, Alan G Isaac wrote:

...
On 3/17/2010 10:16 AM, josef.pktd@gmail.com wrote:

...
numpy.resize(a, new_shape) Return a new array with the specified shape.

If the new array is larger than the original array, then the new array is filled with repeated copied of a. Note that this behavior is different from a.resize(new_shape) which fills with zeros instead of repeated copies of a.

Yes indeed. Sorry, I must have scrolled the help without realizing it, and this part was at the top.

So my follow up: why is this desirable/necessary? (I find it surprising.)

IIRC, it behaved that way in Numeric.

How would people feel about unifying the function vs. the method behavior ? One could add an addition option like `repeat` or `fillZero`. One could (at first !?) keep opposite defaults to not change the current behavior. But this way it would be most visible and clear what is going on. Regards, Sebastian Haase

Gael Varoquaux

1:58 a.m.

On Thu, Mar 18, 2010 at 09:56:21AM +0100, Sebastian Haase wrote:

...

How would people feel about unifying the function vs. the method behavior ? One could add an addition option like `repeat` or `fillZero`.

You mean fill_zero... Sorry, my linter went off. :) Gaël

Alan G Isaac

19 Mar 19 Mar

9:38 a.m.

New subject: np.resize differs from ndarray.resize

On 3/18/2010 4:56 AM, Sebastian Haase wrote:

...

How would people feel about unifying the function vs. the method behavior ? One could add an addition option like `repeat` or `fillZero`. One could (at first !?) keep opposite defaults to not change the current behavior. But this way it would be most visible and clear what is going on.

The current situation is confusing. I therefore hope that one of two things will happen. - unification (probably with a `fill` option, with a default of None that produces the function's behavior but that can be give any numerical value (not just 0)), or - eliminate the method Thanks, Alan

5149

Age (days ago)

5151

Last active (days ago)

List overview

Download

7 comments

6 participants

participants (6)

Alan G Isaac
Charles R Harris
Francesc Alted
Gael Varoquaux
josef.pktd＠gmail.com
Sebastian Haase

bug in ndarray.resize?

Alan G Isaac

josef.pktd＠gmail.com

Alan G Isaac

Charles R Harris

Francesc Alted

Sebastian Haase

Gael Varoquaux

Alan G Isaac

tags

participants (6)