Well, yes and no. Remember that NumPy (I am also a user) does most of its hard work in C. That is why you use a NumPy array and not a normal Python array. It also depends on what you are doing with your data. If you have an immutable list and want to map a (pure) function over it, then that can be trivially parallelized in Haskell. Iterating over an array modifying each element in place is much harder. Remember that lists are always fast for sequential read-only access. Cheers, G On 11 Sep 2010, at 17:28, Henry Olders wrote:
Hi, Lorenzo!
I am also a Haskell beginner, coming from Python, where I also used numpy for speed.
In my python applications, I used lists and list operations (especially list comprehensions) which I found an elegant way of doing things. A numpy array was simply, for my needs, a list of lists.
More experienced users can correct me if I'm wrong, but there are some fundamental differences between Haskell and Python which might influence your choices of lists vs arrays. Haskell, being primarily a compiled language, should give you a speed advantage over interpreted python when you're doing basically the same operations, so you may not need (at least initially) to look for ways to improve speed. My approach basically in any language was to first get the program to work correctly, and to look for speed optimizations only as a secondary step if the program was too slow, and then only for the processes which were the speed bottlenecks.
The second issue is that in Haskell, lists are mutable, whereas arrays are (mostly) immutable. Generally, operations which need to work on every element of a data structure will go faster (and take less memory) if they do not result in the creation of a copy of the data structure, as they need to do for immutable data.
So my tendency would be to look for ways of doing things in Haskell which make use of lists (including lists of lists), and resort to arrays/matrices only for operations which cannot be decomposed into list operations.
Henry
On 2010-09-11, at 10:09 , Lorenzo Isella wrote:
Dear All, The recent feedback I got on the mailing list led me to think about the data structure I need for my computations and array manipulations (loosely speaking, let us say that I need indexing and slicing tables of data). Coming from Python, I am a bit confused: let me say that in my Python scripts I almost never use lists, but rather NumPy arrays. In that case, it is an easy choice (almost every decent software for numerics/visualization etc... in Python relies explicitly or implicitly on NumPy). On top of that, NumPy is fast, has a lot of inbuilt functions and interfaces nicely with SciPy, matplotlib, Mayavi2 etc... It seems to me (please correct me if I am mistaken) that the situation in Haskell is a bit more 'open' to choices. At least I think so when I look at
http://en.wikibooks.org/wiki/Haskell/Hierarchical_libraries/Arrays
I may want to drop lists at some point for performance reasons but also because in my work I really have tables of numbers I find convenient to think of as arrays/matrices (again, loosely speaking, I mean matrices as arrays + linear algebra like taking the inverse, the determinant and so on). Bottom line, I would need a data type that (1) is decently fast (OK, there is more than performance to scripting, but you see my point) (2) allows slicing/indexing (e.g. take 3rd row, second column, flatten it out, take every element larger than 34 in a row, find its unique elements, sort them etc...) without having to re-invent the wheel myself. This is more on the manipulation side than the linear algebra. As you can see, I would like to be able to find something similar to the very useful functions in Data.List for an array. (3) it would be nice if these data type could have either integers of real numbers as entries. If the original data is a made up of integer numbers, conversion to real number is always suspicious (I have horrible memories (at least in other languages) of numbers that were equal as integers and no longer when read as real numbers because of one last digit changing...which can give you a headache in some cases. (4) again it would be nice if I could feed these arrays/vectors to tools that take integrals, derivatives, inverse etc...
I am considering for this reason several possibilities
http://www.haskell.org/tutorial/arrays.html http://users.skynet.be/jyp/html/base/Data-Array.html http://code.haskell.org/hmatrix/hmatrix.pdf
hmatrix is probably what I am looking for (linear algebra, interface to gnuplot for plotting, ODE solver etc...) but there is no possibility of using arrays of integers, so I am concerned about using it to read and compare data files filled with integers where I check if certain entries are equal or not. Also I wonder if I can find any extra documentation other than the well written tutorial (which explains a lot, but cannot do everything in less than 30 pages and I would have plenty of questions about array manipulations there).
Any suggestion/clarification is more than welcome. Cheers
Lorenzo
_______________________________________________ Beginners mailing list Beginners@haskell.org http://www.haskell.org/mailman/listinfo/beginners
_______________________________________________ Beginners mailing list Beginners@haskell.org http://www.haskell.org/mailman/listinfo/beginners