- Numeric - Haskell.org

ANN: set-cover with new examples for Nonogram, Pangram, Mastermind, Lonpos pyramid
by Henning Thielemann 19 Aug '15

19 Aug '15

The package 'set-cover' solves exact set cover problems: https://en.wikipedia.org/wiki/Set_cover_problem I have added new examples: Nonogram, Pangram, Mastermind, Lonpos pyramid.

1 0

ANN: apportionment
by Henning Thielemann 19 Aug '15

19 Aug '15

I have uploaded a new package to Hackage implementing two methods for rounding a set of numbers while maintaining their sum. Useful applications are elections of parliaments and rounding percentages. E.g. Prelude> Apportionment.largestRemainderScaled 100 [1,2,3 :: Double] [17,33,50] Prelude> Apportionment.largestRemainderScaled 100 [1,1,1 :: Double] [34,33,33] The package is very simple, yet useful. It's plain Haskell 98. https://hackage.haskell.org/package/apportionment

1 0

ix-shapable vs. fft
by Henning Thielemann 23 Jul '15

23 Jul '15

I like to discuss some issues about the Haskell package 'fft' package with you. I am thinking for long about ix-shapable and its use in 'fft'. I find it pretty counter-intuitive that FFT.dftRC transforms the outermost dimension of an array and that the dimension list passed to FFT.dftRCN counts from the outermost to the innermost dimension. I guess that the reason is how Shapable methods convert between tuples and lists. In a tuple (a,b,c), 'a' is the zeroth element of the tuple but as an index it counts the outermost dimension. Shapable methods 'sShape' and 'sBounds' convert that tuple to the list [a,b,c] and back. Of course, [a,b,c]!!0 == a. With the working of Shapable in mind, the current convention of counting dimensions in FFT.dftRCN makes sense but most of the time I only use a certain tuple type like (Int,Int,Int) and do not care about its Shapable instance. I do not spell Shapable in the source code, at all. Other problems are: Shapable instances are overly strict in that they allow only tuples that consist entirely of Ints. But e.g. it would be no problem to have the index type (Char, Int, Integer) if we only transform the middle dimension. Additionally the current Shapable instances for tuples require FlexibleInstances, which I find ugly and unnecessary. On the other hand FFT.dftRCN lacks type safety: You can easily specify non-existent dimensions for transformation. The last problem can be overcome by replacing the list of transform dimensions by a Bool tuple. How about a call like this: FFT.dftRCN (False, True, False) array3d ? A fine solution is not as simple as a tuple of Bools, but here is something close to it: ~~~~ import Data.Complex (Complex) import Data.Monoid (Monoid, (<>)) data IndexMagic = InsertMagicHere data Dim ix = Dim IndexMagic instance Monoid IndexMagic where keep :: Dim ix keep = Dim InsertMagicHere trans :: Dim Int trans = Dim InsertMagicHere mask2 :: Dim ix0 -> Dim ix1 -> Dim (ix0,ix1) mask2 (Dim op0) (Dim op1) = Dim (op0 <> op1) mask3 :: Dim ix0 -> Dim ix1 -> Dim ix2 -> Dim (ix0,ix1,ix2) mask3 (Dim op0) (Dim op1) (Dim op2) = Dim (op0 <> op1 <> op2) dftRCN :: Dim i -> array i a -> array i (Complex a) dftRCN = undefined example :: array (Char, Int, Integer) a -> array (Char, Int, Integer) (Complex a) example = dftRCN (mask3 keep trans keep) ~~~~ IndexMagic would be a dictionary of functions for doing the necessary index processing. With this solution we would not need a Shapable class nor a ix-shapable package, and we stay Haskell 98. For some time I prefered a solution using nested Int dimensions as in 'repa' and 'accelerate'. But maybe the tuple approach sketched above isn't so bad. Since the 'fft' package is in use for a rather long time now I hesitate to change something. I will certainly start a new module providing the same function names with new types.

1 0

Math in Haddock
by Dominic Steinitz 14 Jun '15

14 Jun '15

I thought folks might be interested in a little example of what is now possible in Haddock (assuming my PR is accepted): https://hackage.haskell.org/package/Kalman-0.1.0.1/docs/Kalman.html <https://hackage.haskell.org/package/Kalman-0.1.0.1/docs/Kalman.html> Anyone wanting to live on the edge can build their own haddock and upload their haddocks manually to hackage: https://github.com/idontgetoutmuch/haddock Dominic Steinitz dominic(a)steinitz.org http://idontgetoutmuch.wordpress.com

1 0

Hidden Markov Models
by Dominic Steinitz 30 Mar '15

30 Mar '15

> I am currently working on > http://hub.darcs.net/thielema/hmm-hmatrix > > It does not need log-numbers because it normalizes all temporary results. > This way I can use fast hmatrix operations. Would normalization also be a > solution for other probabilistic models? I assume these are models with finite state spaces? FYI I uploaded a Kalman filter https://github.com/idontgetoutmuch/Kalman which is also a HMM. I feel we should be able to abstract over the Markov transition kernel to describe the model without caring whether the state space is continuous or discrete but that it is about as much thinking I have done on the subject. Dominic Steinitz dominic(a)steinitz.org http://idontgetoutmuch.wordpress.com

1 0

Random Number Generation
by Dominic Steinitz 30 Mar '15

30 Mar '15

FYI and please chip in > Edward Kmett <ekmett <at> gmail.com> writes: > > > * random > > > > We've had some truly excellent work done over the last couple of > > years on how to deal with "splitting" a random number generator in a > > cryptographically sound manner. I spent some time cleaning up a few > > outstanding issues for this package personally over the summer, but > > have not had nearly enough time to devote to the issue of how to > > integrate the outcome of the recent research on splitting, while > > simultaneously caring about performance and soundness. > > I finally managed to get some time to start looking at this. I have > created a ticket here > > https://github.com/haskell/random/issues/25 > > Please add comments and any information you think Carter and I might > find helpful. > > I discovered the package has some benchmarks that are 3+ years old. At > the moment, I am not sure what they are telling me. If anyone has any > background / views please chip in. > > Dominic. Dominic Steinitz dominic(a)steinitz.org http://idontgetoutmuch.wordpress.com

1 0

Re: [Numeric] ggplot for Haskell (was Re: [diagrams] Using charts in blogliteratelyd)
by Dominic Steinitz 23 Feb '15

23 Feb '15

Dominic Steinitz dominic(a)steinitz.org http://idontgetoutmuch.wordpress.com On 23 Feb 2015, at 15:50, Daniel Bergey <bergey(a)teallabs.org> wrote: > On 2015-02-17 at 17:41, Dominic Steinitz <dominic(a)steinitz.org> wrote: >> I think we can already do faceting e.g. >> >> https://idontgetoutmuch.files.wordpress.com/2014/06/044408c77f048f73.png?w=… >> http://www.helsinki.fi/varieng/series/volumes/07/siirtola_et_al/figures/sfi… > > How does Chart handle axes bounds when faceting? This is an example > where I believe ggplot outshines matplotlib. matplotlib does a good job > picking axes bounds & tickmarks for a single plot, but if I make two > subplots, it picks bounds separately for each. To get them aligned on > one or both axes, I need to specify bounds myself. ggplot by default > plots each subplot with matching bounds on the matching axes. I should have been clearer - my apologies. chart does *not* do faceting; I transform multiple charts into diagrams and then lay these out in whatever way is appropriate. I can create each chart with identical axes and so get the ggplot experience. > >> but that uses a package called Beanplot >> (http://www.jstatsoft.org/v28/c01/paper) not ggplot. > > That's a cool graph. I've seen symmetric "violin plots" before, with > the same univariate distribution plotted on each side of the vertical > whiskers. I think I prefer this version, with a different category on > each side. I think to do something like this I need to be able to access the components of a chart and be able to create new sorts of components; I don’t believe the former is possible at the moment although I think the latter is (I recall someone just submitted a pull request to produce histograms).

1 0

ggplot for Haskell (was Re: [diagrams] Using charts in blogliteratelyd)
by Dominic Steinitz 17 Feb '15

17 Feb '15

Hi Daniel, Thanks for your comprehensive reply. > I don't feel like I understand the design space enough to start such a > project, but I'd contribute if someone gets it started, or I'd discuss > it more first. I am pretty sure I don't understand the design space either. > Can you say a bit about what a library needs to be "like ggplot", > rather than like other graphing libraries? Is it mostly about > separating the graph description from the data, and composable > abstractions for the graph to support things like faceting? I think we can already do faceting e.g. https://idontgetoutmuch.files.wordpress.com/2014/06/044408c77f048f73.png?w=… I had a quick look at http://byrneslab.net/classes/biol607/readings/wickham_layered-grammar.pdf I may have misunderstood but I think it has a limited scope and calling it a grammar of graphics suggests in can do more than it actually can. That is not to be critical: it is very powerful just not as powerful as the name suggests. What I wanted to draw was something like this: http://www.helsinki.fi/varieng/series/volumes/07/siirtola_et_al/figures/sfi… but that uses a package called Beanplot (http://www.jstatsoft.org/v28/c01/paper) not ggplot. I actually managed to sketch something that was close to what I wanted using diagrams (not chart). http://i.imgur.com/TeyloXA.png So I feel we are quite close to something that is more powerful than ggplot. But my attempts both at the above and at facets are not very flexible / re-usable. However, I feel I am re-inventing the charts wheel. What I feel I need is access to the chart components not to just be able to create a chart. Maybe we should talk directly to the author of charts. I don't think it is possible to create the above example using charts. A further thing one can't do with charts is annotations. I believe one can do this in ggplot. I think as you suggest, we need composable abstractions. For my example, I created a grid and overlaid the histograms on top of it. I'd like to overlay axes on it as well and a legend. > I've used Chart (in Haskell) and ggplot a little, but mostly have > experience with Python using matplotlib & pandas. The last time I > looked at making an R / pandas environment in Haskell, I got bogged > down in picking a dataframe type. tables[1] & Frames[2] both look > relevant. I've used frames and pandas a bit. As far as I understand it they carry around a description of the data along with the data; ggplot can then make use of this. I don't see it as essential and I wouldn't want "ggplot in Haskell" to depend on either tables or frames Maybe we could start "ggplot in Haskell" by somehow generalizing my feeble examples? Dominic Steinitz dominic(a)steinitz.org http://idontgetoutmuch.wordpress.com On 17 Feb 2015, at 00:02, Daniel Bergey <bergey(a)teallabs.org> wrote: > On 2015-02-16 at 09:15, idontgetoutmuch <idontgetoutmuch(a)googlemail.com> wrote: >> 2. Why does haskell-chart load fonts but diagrams itself doesn't? Does >> diagrams generate fonts on the fly somehow? > > The answer depends on which Diagrams Backend you are using. Backends > which output raster images load the fonts they use. Vector output > formats may get away refering to the font name, and expecting the viewer > to load the font. Some vector backends can access font metrics to > position (eg, center) text. > > I may be wrong about some of the following table: > > | Backend | Embed Font / bitmap | Font Metrics | > |------------+--------------------------+--------------| > | Cairo | depends on output format | Y | > | SVG | N | partial | > | Rasterific | Y | partial | > | Canvas | N | partial | > > I believe Chart loads fonts to provide better text positioning than the > Diagrams Backends currently support. I think there's some low-hanging > fruit, here, notably > https://github.com/diagrams/diagrams-rasterific/issues/21 > now that Rasterific has the needed feature. > >> 3. Has anyone shown any interest in writing something like ggplot using >> diagrams? This would be a fantastic tool but wouldn't be a trivial >> undertaking. > > Quite a few people have expressed interest in *using* such a library. I > haven't heard anyone say they have time to write it, but maybe between > us we do. > > I don't feel like I understand the design space enough to start such a > project, but I'd contribute if someone gets it started, or I'd discuss > it more first. > > Can you say a bit about what a library needs to be "like ggplot", rather > than like other graphing libraries? Is it mostly about separating the > graph description from the data, and composable abstractions for the > graph to support things like faceting? > > I've used Chart (in Haskell) and ggplot a little, but mostly have > experience with Python using matplotlib & pandas. The last time I > looked at making an R / pandas environment in Haskell, I got bogged down > in picking a dataframe type. tables[1] & Frames[2] both look relevant. > > cheers, > bergey > > Footnotes: > [1] http://hackage.haskell.org/package/tables > > [2] https://github.com/acowley/Frames >

1 0

Welcome to the Numeric Haskell List
by Gershom B 15 Feb '15

15 Feb '15

Frist Post! (sorry/notsorry) —gershom

1 0