Hi all, This discussion has brought up a question I have had for a while: Can anyone provide a one-paragraph description of what numarray does that gives it better large-array performance than Numeric? By the way, For what it's worth, what's kept me from switching is the small array performance, and/or the array-creation performance. I don't use very large arrays, but I do use small ones all the time. thanks, -Chris -- Christopher Barker, Ph.D. Oceanographer NOAA/OR&R/HAZMAT (206) 526-6959 voice 7600 Sand Point Way NE (206) 526-6329 fax Seattle, WA 98115 (206) 526-6317 main reception Chris.Barker@noaa.gov

Ralf Juengling wrote:

I took a cursory look at the C API the other day and learned about this capability to process byte-swapped data. I am wondering why this is a good thing to have. Wouldn't it be enough and much easier to drop this feature and instead equip numarray IO routines with the capability to convert to and from a foreign endian to the host endian encoding?

Basically this feature was to allow use of memory mapped data that didn't use the native representation of the processor (also related to supporting record arrays). The details are given in a paper a couple years ago: http://www.stsci.edu/resources/software_hardware/numarray/papers/pycon2003.p df Perry

I haven't followed this discussion in detail but with respect to space for 'descriptors', it would simply be foolish to malloc space for these. The cost is ridiculous. You simply have to decide how big a number of dimensions to allow, make it a clearly findable definition in the sources, and dimension everything that big. Originally when we discussed this we considered 7, since that had been (and for all I know still is) the maximum array dimension in Fortran. But Jim Huginin needed 11 or something like it for his imaging. I've seen 40 in the numarray sources I think. It seems to me that an application that would care about this space (it being, after all, per array object) would be unusual indeed. If I've misunderstood what you're talking about, never mind. (:-> My advice is to make flexibility secondary to performance. It is always possible to layer on flexibility for those who want it.

Paul Dubois wrote:

I haven't followed this discussion in detail but with respect to space for 'descriptors', it would simply be foolish to malloc space for these. The cost is ridiculous. You simply have to decide how big a number of dimensions to allow, make it a clearly findable definition in the sources, and dimension everything that big.

Originally when we discussed this we considered 7, since that had been (and for all I know still is) the maximum array dimension in Fortran. But Jim Huginin needed 11 or something like it for his imaging. I've seen 40 in the numarray sources I think.

Actually, 40 came from Numeric. It may have been reduced to 11, but I'm sure it was 40 at one point. Jim even had a comment in the code to the effect that if someone needed more than 40, he wanted to see the problem that needed that. If people think it is too high, I'd be very happy to reduce it. Perry

Robert Kern <rkern@ucsd.edu> writes:

Travis Oliphant wrote:

...

...

...

4) Object arrays must be supported. This was a bad oversight and an important feature of Numeric arrays.

The current implementation does support them (though in a different way, and generally not as efficiently, though Todd is more up on the details here). What aspect of object arrays are you finding lacking? C-api? I did not see such support when I looked at it, but given the previous comment, I could easily have missed where that support is provided. I'm mainly following up on Konrad's comment that his Automatic differentiation does not work with Numarray because of the missing support for object arrays. There are other applications for object arrays as well. Most of the support needs to come from the ufunc side.

It's tucked away in numarray.objects. Unfortunately for Konrad's application, numarray ufuncs don't recognize that it's being passed an object with the special methods defined, and they won't automatically create 0-D object "arrays". 0-D object arrays will work just fine when using operators (x+y works), but not when explicitly calling the ufuncs (add(x,y) does not work). Both methods work fine for 0-D numerical arrays.

Are the 0-D object arrays necessary for this? The behaviour that Konrad needs is this (highly abstracted): class A: def __add__(self, other): return 0.1 def sin(self): return 0.5 Then:

...

...

a = A() a + a 0.10000000000000001 Numeric.add(a,a) 0.10000000000000001 Numeric.sin(a) 0.5

The Numeric ufuncs, if the argument isn't an array, look for a method of the right name (here, sin) on the object, and call that. You could define a delegate class that does this with something like class MathFunctionDelegate: def __init__(self, fallback=Numeric): self._fallback = fallback def add(self, a, b): try: return a + b except TypeError: return self._fallback.add(a, b) def sin(self, x): sin = getattr(x, 'sin', None) if sin is None: return self._fallback.sin(x) else: return sin(x) ... etc. ... (This could be a module, too. This just allows parameterisation.) In ScientificPython, FirstDerivatives.py has a method of the DerivVar class that looks like this: def sin(self): v = Numeric.sin(self.value) d = Numeric.cos(self.value) return DerivVar(v, map(lambda x,f=d: f*x, self.deriv)) Add something like this to the __init__: self._mathfuncs = MathFunctionDelegate(Numeric) and that sin method becomes def sin(self): v = self._mathfuncs.sin(self.value) d = self._mathfuncs.cos(self.value) return DerivVar(v, map(lambda x,f=d: f*x, self.deriv)) That's not quite perfect, as the user has to use a mathfuncs object also; that's why having Numeric or numarray do the delegation automatically is nice. This would work equally well with numarray (or the math or cmath modules!) replacing Numeric. You could get fancy and be polymorphic: choose the right module to use depending on the type of the argument (Numeric arrays use Numeric, floats use math, etc.). If this was a module instead, you could have registration of types. I'll call this module numpy. Here's a possible (low-level) usage: import numpy import Numeric, numarray, math, cmath from Scientific.Functions import Derivatives numpy.register_type(Numeric.arraytype, Numeric) numpy.register_type(numarray.NumArray, numarray) numpy.register_type(float, math) numpy.register_type(complex, cmath) numpy.register_type(Derivatives.DerivVar, Derivates.derivate_math) numpy.default_constructor(numarray.array) a = numpy.array([1,2,3]) # makes a numarray b = Numeric.array([1,2,3]) # Numeric array print numpy.sin(a), numpy.sin(b) Things to consider with this would be: * how to handle a + b * where should the registering of types be done? (Probably by the packages themselves) * more complex predicates for registering handlers? (to handle subclasses, etc.) etc. Ok, I hope that's not too rambling. But the idea is that neither Numeric nor numarray need to provide the delegation ability. -- |>|\/|< /--------------------------------------------------------------------------\ |David M. Cooke http://arbutus.physics.mcmaster.ca/dmc/ |cookedm@physics.mcmaster.ca

Travis Oliphant

Thanks for the comments that have been made. One of my reasons for commenting is to get an understanding of which design issues of Numarray are felt to be important and which can change. There seems to be this idea that small arrays are not worth supporting. I hope this is just due to time-constraints and not some fundamental idea that small arrays should never be considered with Numarray. Otherwise, there will always be two different array implementations developing at their own pace.

I wouldn't say that we are "hostile" to small arrays. We do only have limited resources and can't do everything we would like. More on this below though.

I really want to gauge how willing developers of numarray are to changing things.

Without going into all the details below, I think I can address this point. I suppose it all depends on what you mean by "how willing developers of numarray are to changing things." If you mean are we open to changes to numarray that speed up small arrays (and address other noted shortcomings). Yes, certainly (so long as they don't hurt the large array issues significantly). If it means we will drop everything and address all these issues immediately ourselves. No, we have other things to do regarding numarray that have higher priority before we can address these things. I would have a very hard time justifying the effort when there are other things needed by STScI more. We would love it if others could address them sooner though. More on related issues below.

...

...

1) Are there plans to move the nd array entirely into C? [...] I do not think it would be difficult at this point to move it all to C and then make future changes there (you can always call pure Python code from C). With the structure in place and some experience behind you, now seems like as good a time as any. Especially, because now is a better time for me than any... I like what numarray is doing by not always defaulting to ints with the maybelong type. It is a good idea.

I hope that is true, but we've found doing moving thing to C a bigger effort than we would like. I'd like to be proved wrong by someone who can tackle sooner than we can.

...

...

2) Why is the ND array C-structure so large? Why are the dimensions and strides array static? Why can't the extra stuff that the fancy arrays need be another structure and the numarray C structure just extended with a pointer to the extra stuff?

When Todd moved NDArray into C, he tried to keep it simple. As such, it has no "moving parts." We think making dimensions and strides malloc'ed rather than static would be fairly easy. Making the "extra stuff"

...

variable is something we can look at.

But allocating dimensions and strides when needed is not difficult and it reduces the overhead of the ndarray object. Currently, that overhead seems extreme. I could be over-reacting here, but it just seems like it would have made more sense to expand the array object as little as possible to handle the complexity that you were searching for. It seems like more modifications were needed in the ufunc then in the arrayobject.

I'm not convinced that this is a big issue, but we have no objection to someone making this change. But it falls well below small array performance in priority for us.

...

The bottom line is that adding the variability adds complexity and we're not sure we understand the storage economics of why we would doing it. Numarray was designed, first and foremost, for large arrays.

Small arrays are never going to disappear (Fernando Perez has an excellent example) and there are others. A design where a single pointer not being NULL is all that is needed to distinguish "simple" Numeric-like arrays from "fancy" numarray-like arrays seems like a great way to make sure that

I won't quarrel with that (but I'm not sure what you are suggesting in the bigger picture).

On another fundamental note, numarray is being sold as a replacement for Numeric. But, then, on closer inspection many things that Numeric does well, numarray is ignoring or not doing very well. I think this presents a certain amount of false advertising to new users, who don't understand the history. Most of them would probably never need the fanciness that numarray provides and would be quite satisfied with Numeric. They just want to know what others are using. I think it is a disservice to call numarray a replacement for Numeric until it actually is. It should currently be called an "alternative implementation" focused on large arrays. This (unintentional) slight of hand that has been occurring over the past year has been my biggest complaint with numarray. Making numarray a replacement for Numeric means that it has to support small arrays, object arrays, and ufuncs at least as well as but preferably better than Numeric. It should also be faster than Numeric whenever possible, because Numeric has lots of potential optimizations that have never been applied. If numarray does not do these things, then in my mind it cannot be a replacement for Numeric and should stop being called that on the numpy web site.

It distresses me to be accused of false advertising. We were pretty up front at the beginning of the process of writing numarray that the approach we would be taking would likely mean slower small array performance. There were those (like you and Eric that expressed concern about that), but it wasn't at all clear what the consensus was regarding how much it could change and be acceptable. (I recall at one point when IDL was ported from Fortran to C which resulted in a factor of 2 overall slowdown in speed. People didn't accuse RSI of providing something that wasn't a replacement for IDL.) The fact was that at the time we started, several thought that backward compatibility wasn't that important. We didn't even try at the beginning to make the C-API the same. At the start, there was no claim that numarray would be an exact replacement for Numeric. (And I didn't hear huge objections at the time on the point and some that actually encouraged a break with how Numeric did things.) Much of the attempts to provide backward compatiblity have come well after the first implementations. We have strove to provide the full functionality of what Numeric had as we went to version 1.0. Sure, there are some holes for object arrays. So the issue of whether numarray is a replacement or not seems to be arguing over what the intent of the project was. Paul Dubois wrote the numpy page that make that reference, and sure, I didn't object to it (But why didn't you at the time? It's been there a long time, and the goals and direction of numarray have been quite visible for a long time. This wasn't some dark, secret project. Many of the things you are complaining about have been true for some time.) If people want to call numarray an alternative implementation, I'm fine with that. It was a replacement in our case. If we didn't develop it, we likely wouldn't be using Python in the full sense that we are now. Numeric wasn't really an option. At the time, many supported the idea of a reimplementation so it seemed like a good opportunity to add what we needed and do that. Obviously, we misread the importance of small array performance for a significant part of the community. (But I keep saying, if small array peformance is really that important, it would seem to me that much bigger wins are available as Fernando mentioned) It's been clear for a better part of a year that it would be a long time before there was any sort of unification between the two. That distressed me as I'm sure it did you. So some sort of useful sharing of libraries and packages seemed like the obvious way to go. In more specialized areas, there would be some divergence (e.g., we have dependencies on record arrays that we just can't provide in Numeric). I can no longer justify sinking many more months of work into numarray for issues of no value to STScI (other than the hope that it would convince others to switch, which isn't clear at all that it would). We need to move towards providing a lot of the tools that are available for Numeric. I can justify that work. The current situation is far from ideal (Paul called it "insane" at scipy if you prefer more colorful language). What we have are two camps that cannot afford to give up the capabilities that are unique to each version. But with most of the C-API compatable, and a way of coding most libraries (except for Ufuncs) to be compatible with both, we certainly can improve the situation. If you can help remove the biggest obstacle, small array performance, so that we could unify the two I would be thrilled, but most of the effort can't come from us, at least not in the near term (next year). We can help at some level. [...]

I never really understood the "code is too complicated" argument

You lost me on this one. You mean the complaint that it was too complicated in Numeric way back?

anyway. I was just wondering if there is some support for reducing the number of source code files, or reorganizing them a bit.

Yes, I'd say that this has relatively high priority. It would be nice to have feedback and advice on how to do this best.

...

...

4) Object arrays must be supported. This was a bad oversight and an important feature of Numeric arrays.

The current implementation does support them (though in a different way, and generally not as efficiently, though Todd is more up on the details here). What aspect of object arrays are you finding lacking? C-api?

I did not see such support when I looked at it, but given the previous comment, I could easily have missed where that support is provided. I'm mainly following up on Konrad's comment that his Automatic differentiation does not work with Numarray because of the missing support for object arrays. There are other applications for object arrays as well. Most of the support needs to come from the ufunc side.

I think Robert Kern pointed to the issue in a subsequent message.

...

...

Again, thanks to the work that has been done. I'm really interested to see if some of these modifications can be done as in my mind it will help the process of unifying the two camps.

I'm glad to see that you are taking a look at it and welcome the comments and any offers of help in improving speed.

I would be interested in helping if there is support for really making numarray a real replacement for Numeric, by addressing the concerns that I've outlined. As stated at the beginning, I'm really just looking for how receptive numarray developers would be to the kinds of changes I'm talking about: (1) reducing the size of the array structure, (2) moving the ndarray entirely into C, (3) improving support for object arrays, (4) improving ufunc API support.

I'm not exactly sure what you mean by 4). If you mean having a compatible api to numeric, that seem like a lot of work since the way ufuncs work in numarray is quite different. But you may mean something else. Perry

On Jan 19, 2005, at 7:31 AM, konrad.hinsen@laposte.net wrote:

...

The current situation is far from ideal (Paul called it "insane" at scipy if you prefer more colorful language). What we have are two camps that cannot afford to give up the capabilities that are unique to each version. But with most of the C-API compatable, and a way of coding most libraries (except for Ufuncs) to be compatible with both, we certainly can improve the situation.

I am not sure that compatibility is really the main issue. In the typical scientific computing installation, NumPy and numarray are building blocks. Some people use them without even being aware of them, indirectly through other libraries.

In a building-block world, two bricks should be either equivalent or be able to coexist. The original intention was to make NumPy and numarray equivalent, but this is not what they are at the

Just to clarify, the intention to make them equivalent was not originally true (and some encouraged the idea that there be a break with Numpy compatibility). But that has grown to be a much bigger goal over time.

moment. But they do not coexist very well either. While it is easy to install both of them, every library that builds on them uses one or the other (and to make it worse, it is not always easy to figure out which one is used if both are available). Sooner or later, anyone who uses multiple libraries that are array clients is going to have a compatibility issue, which will probably be hard to understand because both sides' arrays look so very similar.

No doubt that supporting both introduces more work, but for the most part, I think that with the exception of some parts(namely ufunc C-api), it should be possible write a library that supports both with little conditional code. That does mean not using some features of numarray, or depending some of the different behaviors of Numeric (e.g., scalar coercion rules), so that requires understanding the subsets to use. And that does cost. But one doesn't need to have two separate libraries. In such cases I'm hoping there is no need to mix different flavors of arrays. You either use Numeric arrays consistently or numarrays consistently. And if the two can be unified, then this will just be a intermediate solution. Perry

I'd like to clarify our position on this a bit in case previous messages have given a wrong or incomplete impression. 1) We don't deny that small array performance is important to many users. We understand that. But it generally isn't important for our projects, and in the list of things to do for numarray, we can't give it high priority this year. We have devoted resources to this issue in the past couple of years (but without sufficient success to persuade many to switch for that reason alone), and it is hard to continue to put much more resources into this not knowing whether it will be enough of an improvement to satisfy those that really need it. 2) This doesn't mean that we don't think it isn't important to add as soon as it can be done. That is, we aren't trying to prevent such improvements from being made. 3) We hope that there are people out there for which this is important who would like to see a numarray/Numeric unification, have some experience with the internals of one or the other (or are willing to learn), and are willing to devote the time to help make numarray faster (if you can rewrite everything from scratch and satisfy both worlds, that would make just as happy :-). 4) We are willing to help in the near term as far as helping explain how things currently work, where possible improvements can be made, helping in design discussions, reviewing proposed or actual changes, and doing the testing and integration of such changes. 5) But the onus of doing the actual implementation can't be on us for reasons I've already given. But besides those I think it is important that whoever does this should have a strong stake in the success of this (i.e., the performance improvements are important for their projects). Perry

On 18.01.2005, at 19:26, Travis Oliphant wrote:

On another fundamental note, numarray is being sold as a replacement for Numeric. But, then, on closer inspection many things that Numeric does well, numarray is ignoring or not doing very well. I think this presents a certain amount of false advertising to new users, who don't understand the history. Most of them would probably never need the fanciness that

I agree with that. I regularly get questions from people who download my code and then wonder why it "still" uses NumPy instead of the "newer" numarray. The reason is that my code has nothing to gain from numarray, as it uses many small and few if any very large arrays. I have no problem explaining that, but the fact that the question arises shows that there is a wrong perception by many newcomers of the relation between NumPy and numarray.

comment, I could easily have missed where that support is provided. I'm mainly following up on Konrad's comment that his Automatic differentiation does not work with Numarray because of the missing support for object arrays. There are other applications for object arrays as well. Most of the

While I agree that object arrays are useful, they have nothing to do with the missing feature that I mentioned recently. That one concerns only ufuncs. In NumPy, they use a method call when presented with an object type they cannot handle directly. In numarray, they just produce an error message in that case. Returning to object arrays, I have used them occasionally but never in any of my public code, because there have been lots of minor bugs concerning them in all versions of NumPy. It would be nice if numarray could do a better job there. Konrad. -- ------------------------------------------------------------------------ ------- Konrad Hinsen Laboratoire Leon Brillouin, CEA Saclay, 91191 Gif-sur-Yvette Cedex, France Tel.: +33-1 69 08 79 25 Fax: +33-1 69 08 82 61 E-Mail: hinsen@llb.saclay.cea.fr ------------------------------------------------------------------------ -------