[Numpy-discussion] proposal: new logspace without the log in the argument

[.]

I've suggested a new function similar to logspace, but where you specify the start and stop points directly instead of using log(start) and base arguments:

https://github.com/numpy/numpy/issues/7255
https://github.com/numpy/numpy/pull/7268

[Joseph Fox-Rabinovitz]

I like the idea, as long as we all remain aware of the irony of having
a "log" spacing for a function named "lin"space.

-Joe

- any better ideas for naming it than "geomspace"? It's really too bad
that the 'logspace' name is already taken.

geomspace() is a perfectly cromulent name, IMO.

- I guess the alternative interface might be something like
np.linspace(start, stop, steps, spacing="log")
what do people think?

In a new function not named `linspace()`, I think that might be fine. I do
occasionally want to swap between linear and logarithmic/geometric spacing
based on a parameter, so this doesn't violate the van Rossum Rule of
Function Signatures.
--
Robert Kern
_______________________________________________
NumPy-Discussion mailing list
https://mail.scipy.org/mailman/listinfo/numpy-discussion

[Peter Creasey]

- any better ideas for naming it than "geomspace"? It's really too bad
that the 'logspace' name is already taken.
- I guess the alternative interface might be something like
np.linspace(start, stop, steps, spacing="log")
what do people think?
-n

You’ve got to wonder how many people actually use logspace(start,
stop, num) in preference to 10.0**linspace(start, stop, num) - i.e. I
prefer the latter for clarity, and if I wanted performance I’d be
prepared to write something more ugly.

I don’t mind geomspace(), but if you are brainstorming

linlogspace(start, end) # i.e. ‘linear in log-space’

is ok for me too.

Peter

[Robert Kern]

With respect to geomspace proposals: instead of specifying start and end

values and the number of points I'd like to have an option where I can set
the start and end points and the ratio. The function would then work out
the correct number of points to get closest to the end value.

E.g. geomspace(start=1, finish=2, ratio=1.03)
The first entries would be 1.0, 1.03, 1*1.03**2, etc.
I have a requirement for the correct ratio between the points, and it's a

right bind having to calculate the exact number of points needed.

At the risk of extending the twisty little maze of names, all alike, I
would probably call a function with this signature geomrange() instead. It
is more akin to arange(start, stop, step) than linspace(start, stop,
num_steps).

--
Robert Kern

[Robert Kern]

In a new function not named `linspace()`, I think that might be fine. I

do occasionally want to swap between linear and logarithmic/geometric
spacing based on a parameter, so this

doesn't violate the van Rossum Rule of Function Signatures.

Would such a new function correct the apparent mistake (?) of
`linspace` including the endpoint by default?
Or is the current API justified by its Matlab origins?
(Or have I missed the point altogether?)

The last, I'm afraid. Different use cases, different conventions. Integer
ranges are half-open because that is the most useful convention in a
0-indexed ecosystem. Floating point ranges don't interface with indexing,
and the closed intervals are the most useful (or at least the most common).

If this query is annoying, please ignore it. It is not meant to be.

The same for my answer.

--
Robert Kern

[Joseph Fox-Rabinovitz]

If the author is willing, I'd say both functions are useful. The "geom" prefix is very fitting.
- Joe

With respect to geomspace proposals: instead of specifying start and end values and the number of points I'd like to have an option where I can set the start and end points and the ratio. The function would then work out the correct number of points to get closest to the end value.
E.g. geomspace(start=1, finish=2, ratio=1.03)
The first entries would be 1.0, 1.03, 1*1.03**2, etc.
I have a requirement for the correct ratio between the points, and it's a right bind having to calculate the exact number of points needed.

At the risk of extending the twisty little maze of names, all alike, I would probably call a function with this signature geomrange() instead. It is more akin to arange(start, stop, step) than linspace(start, stop, num_steps).
--
Robert Kern

[.]

What about this API? You specify the start point, ratio, and number of
points.

http://spacepy.lanl.gov/doc/autosummary/spacepy.toolbox.geomspace.html

On Fri, Feb 19, 2016 at 7:10 AM, Andrew Nelson andyfaff-at-gmail.com |numpy

With respect to geomspace proposals: instead of specifying start and end
values and the number of points I'd like to have an option where I can set
the start and end points and the ratio. The function would then work out
the correct number of points to get closest to the end value..
E.g. geomspace(start=1, finish=2, ratio=1.03)
The first entries would be 1.0, 1.03, 1*1.03**2, etc.
I have a requirement for the correct ratio between the points, and it's a
right bind having to calculate the exact number of points needed.
_______________________________________________
NumPy-Discussion mailing list
https://mail.scipy.org/mailman/listinfo/numpy-discussion

[Chris Barker]

Would such a new function correct the apparent mistake (?) of
`linspace` including the endpoint by default?
Or is the current API justified by its Matlab origins?

I don't think so -- we don't need no stinkin' Matlab !

But I LIKE including the endpoint in the sequence -- for the common use
cases, it's often what you want, and if it didn't include the end point but
you did want that, it would get pretty ugly to figure out how to get what
you want.

On the other hand, if I had it to do over, I would have the count specify
the number of intervals, rather than the number of items. A common cae may
be: values from zero to 10 (inclusive), and I want ten steps:

In [19]: np.linspace(0, 10, 10)
Out[19]:

array([ 0. , 1.11111111, 2.22222222, 3.33333333,
4.44444444, 5.55555556, 6.66666667, 7.77777778,
8.88888889, 10. ])

HUH? I was expecting [0,1,2,3 ....] (OK, not me, this isn't my first
Rodeo), so now I need to do:
In [20]: np.linspace(0, 10, 11)
Out[20]: array([ 0., 1., 2., 3., 4., 5., 6., 7., 8., 9.,
10.])

This gets uglier if I know what "delta" I want:

In [21]: start = 0.0; end = 9.0; delta = 1.0
In [24]: np.linspace(start, end, (end-start)/delta)
Out[24]: array([ 0. , 1.125, 2.25 , 3.375, 4.5 , 5.625, 6.75 ,
7.875, 9. ])

oops!

In [25]: np.linspace(start, end, (end-start)/delta + 1)
Out[25]: array([ 0., 1., 2., 3., 4., 5., 6., 7., 8., 9.])

But in any case, there is no changing it now.

-CHB
--
Christopher Barker, Ph.D.
Oceanographer

Emergency Response Division
NOAA/NOS/OR&R [phone] voice
7600 Sand Point Way NE [phone] fax
Seattle, WA 98115 [phone] main reception

***@noaa.gov