comp.lang.idl-pvwave archive: archive

Home » Public Forums » archive » Beginner, Question

Show: Today's Messages :: Show Polls :: Message Navigator
E-mail to friend

Re: Beginner, Question [message #91829 is a reply to message #91824]

Thu, 03 September 2015 04:43

Fabzi
Messages: 305
Registered: July 2010

Senior Member

Thanks Paulo, I totally agree with your analysis. Having tried out
Python myself for a new project, I will add a few things:

1) awkwardness of Numpy

I totally agree with you and dislike the mix between numpy's ndarray's
and python lists. Specially the syntax overhead is not exactly aesthetic
(eg a = np.array([1,2,3]) instead of a simple a = [1, 2, 3]). However,
I've come to really appreciate some of Numpy's broadcasting syntax,
which simplifies some operations, for example:

>>> a = np.zeros((3, 4))
>>> a += [1, 2, 3, 4]
>>> a
array([[ 1., 2., 3., 4.],
[ 1., 2., 3., 4.],
[ 1., 2., 3., 4.]])

To be compared to IDL:

IDL> a = FLTARR(4, 3)
IDL> a += [1, 2, 3, 4] # (FLTARR(3)+1)
IDL> a
1.0000000 2.0000000 3.0000000 4.0000000
1.0000000 2.0000000 3.0000000 4.0000000
1.0000000 2.0000000 3.0000000 4.0000000

And there are some other advantages (e.g the ellipsis syntax:
a[1:3, ...] which helps you not to bother about variable dimensions any
more)

2) Debugger

I agree with you again. One thing that surprised me is how slow Python
debuggers are, and that they require a separate running mechanism. IDL
in turn has no formal "debug mode" (just put a breakpoint and wait) and
is running at the exact same speed. Does anyone know why the two
debuggers are so different?

3) Python third party libraries

The major reason for choosing python for my new project (aside of the
license issue) is simply that I spared a HUGE amount of work, simply
because the things we badly needed were already there. I wont give a
full list here but basically, after a year using libraries such as
Pandas, Shapely, or Scikit-Learn, going back to IDL for serious data
crunching is quite frustrating...

I think that the recent efforts of the IDL devs to make the syntax more
flexible are the right ones, and the idea to allow python - IDL bindings
is really great. But IMO the next steps should be a revisiting of some
unsatisfying basic functionalities:
- the handling of time: IDL's floating julian days are not adapted to
today's standards, and plotting timeseries still is a pain.
- it would be great to have indexed and labeled arrays such as in Pandas
and xray...
- it would be much nicer to be able to do:

IDL> a = list([1, 2, 3], DICTIONARY('a', 1, 'b', 2))
IDL> a[0][1]
2
IDL> a[1].b
2

instead of the current syntax:

IDL> (a[0])[1]
2
IDL> (a[1])['b']
2

Cheers,

Fabien

On 09/02/2015 10:22 PM, Paulo Penteado wrote:
> snip

Report message to a moderator

[Message index]

		Beginner, Question By: justdoit0429 on Mon, 24 August 2015 21:49
		Re: Beginner, Question By: Jeremy Bailin on Tue, 25 August 2015 07:06
		Re: Beginner, Question By: chris_torrence@NOSPAM on Wed, 26 August 2015 08:00
		Re: Beginner, Question By: penteado on Wed, 02 September 2015 13:22
		Re: Beginner, Question By: Fabzi on Thu, 03 September 2015 04:43
		Re: Beginner, Question By: Yngvar Larsen on Thu, 03 September 2015 05:31
		Re: Beginner, Question By: Fabzi on Thu, 03 September 2015 05:32
		Re: Beginner, Question By: Yngvar Larsen on Thu, 03 September 2015 05:49
		Re: Beginner, Question By: Jim Pendleton on Thu, 03 September 2015 06:04
		Re: Beginner, Question By: penteado on Thu, 03 September 2015 14:11
		Re: Beginner, Question By: Jeremy Bailin on Thu, 03 September 2015 10:28

Previous Topic:	FG PLOT options HISTOGRAM/STAIRSTEP but for horizontal display?
Next Topic:	round up

-=] Back to Top [=-

[ Syndicate this forum (XML) ] [

] [

]

Current Time: Fri Oct 10 02:38:22 PDT 2025

Total time taken to generate the page: 1.12388 seconds