Functional unparsing

Functional unparsing

[Andrey Riabushenko]

Hi ocaml developers,

I am currently developing a library for multivariate regressions including 
linear, nonlinear, generalized, weighted and so on. In my opinion most the 
convenient interface is the one described further and is based on functional 
unparsing.

Something like that:

Stats.linear_regression "y ~ x1 exp(x2) log(x3) x3^2" 
Returns float -> float -> float -> float -> regression_result = <fun>

Stats.generalized_regression "log(y) ~ x1 x^2 log(x3) " 
Returns float -> float -> float -> float -> regression_result = <fun>

For time series:
Stats.ts_regression "y ~ ARMA(5,3)"
Stats.ts_regression "y ~ ARMA(2,2) GARCH(2,1)"


If want to ask you two questions. 
1. Do you find such interface convenient? Critique is welcome. If you have 
better idea, please tell me. I will make all publicly available.

2. The second question regarding function unparsing. I haven't used this 
technique before. Are there some docs, blog articles, descriptions and etc? 
The only relevant documentation I have found is printf.ml :). Might someone 
have a minimal working example to demonstrate?

Re: [Caml-list] Functional unparsing

[Daniel Bünzli]

Le 15 avr. 09 à 22:41, Andrey Riabushenko a écrit :

> Something like that:
>
> Stats.linear_regression "y ~ x1 exp(x2) log(x3) x3^2"
> Returns float -> float -> float -> float -> regression_result = <fun>

It won't work with control strings as plain strings -- what would the  
type of Stats.linear_regression be ? Functional unparsing solves the  
problem by using values of a particular type and combinators to  
represent the control string (you can actually see it as directly  
manipulating the abstract syntax tree of the corresponding control  
string).

Have a look at Olivier Danvy's paper [1]. You may also be interested  
in language embedding in general, see this message [2] for a reference  
and an example.

Best,

Daniel


[1] http://www.brics.dk/RS/98/12/BRICS-RS-98-12.pdf
[2] http://caml.inria.fr/pub/ml-archives/caml-list/2008/03/8bc5f07678134c77037c1a2feedd4d90.en.html

Re: [Caml-list] Functional unparsing

[Hezekiah M. Carty]

On Wed, Apr 15, 2009 at 4:41 PM, Andrey Riabushenko <cdome@bk.ru> wrote:
> Something like that:
>
> Stats.linear_regression "y ~ x1 exp(x2) log(x3) x3^2"
> Returns float -> float -> float -> float -> regression_result = <fun>
...
> 2. The second question regarding function unparsing. I haven't used this
> technique before. Are there some docs, blog articles, descriptions and etc?
> The only relevant documentation I have found is printf.ml :). Might someone
> have a minimal working example to demonstrate?

The Batteries Print module and associated syntax extension may be a
useful base for implementing something similar to what you are
proposing:

http://git.ocamlcore.org/cgi-bin/gitweb.cgi?p=batteries/batteries.git;a=blob_plain;f=src/core/extlib/print.ml;hb=HEAD
and
http://git.ocamlcore.org/cgi-bin/gitweb.cgi?p=batteries/batteries.git;a=tree;f=src/syntax/pa_strings;hb=HEAD

This allows for syntax like:
Print.printf p"This is a list of integers: %{int list}"
to return:
int list -> unit = <fun>

Hez

-- 
Hezekiah M. Carty
Graduate Research Assistant
University of Maryland
Department of Atmospheric and Oceanic Science

Re: [Caml-list] Functional unparsing

[Andrey Riabushenko]

> The Batteries Print module and associated syntax extension may be a
> useful base for implementing something similar to what you are
> proposing:
>
> http://git.ocamlcore.org/cgi-bin/gitweb.cgi?p=batteries/batteries.git;a=blo
>b_plain;f=src/core/extlib/print.ml;hb=HEAD and
> http://git.ocamlcore.org/cgi-bin/gitweb.cgi?p=batteries/batteries.git;a=tre
>e;f=src/syntax/pa_strings;hb=HEAD
>
> This allows for syntax like:
> Print.printf p"This is a list of integers: %{int list}"
> to return:
> int list -> unit = <fun>
>

Thank you, this one was much more helpful then standard printf.ml. This one is 
much shorter and easier to understand.

Re: [Caml-list] Functional unparsing

[Cedric Auger]

Andrey Riabushenko a écrit :
> Hi ocaml developers,
>
> I am currently developing a library for multivariate regressions including 
> linear, nonlinear, generalized, weighted and so on. In my opinion most the 
> convenient interface is the one described further and is based on functional 
> unparsing.
>
> Something like that:
>
> Stats.linear_regression "y ~ x1 exp(x2) log(x3) x3^2" 
> Returns float -> float -> float -> float -> regression_result = <fun>
>
> Stats.generalized_regression "log(y) ~ x1 x^2 log(x3) " 
> Returns float -> float -> float -> float -> regression_result = <fun>
>
> For time series:
> Stats.ts_regression "y ~ ARMA(5,3)"
> Stats.ts_regression "y ~ ARMA(2,2) GARCH(2,1)"
>
>
> If want to ask you two questions. 
> 1. Do you find such interface convenient? Critique is welcome. If you have 
> better idea, please tell me. I will make all publicly available.
>   
I don't know what is really itended, but  I think implementing something 
like
" let double x = 2 *. x in
  Stats.linear_regression "y ~ double x1""
  will be painfull, since you need to parse
  the string according to the environnement.

 The solution
 " let double x = 2 *. x in
  Stats.linear_regression "y ~ %1 x1" double"
closer to printf and format %1 meaning "float -> float" is easier to 
implement,
but less readable.

If you don't require custom function, what you think should do the trick 
and seems to be manageable, but maybe that will be awfull to repeat many 
times the same function in case of use in different places.
> 2. The second question regarding function unparsing. I haven't used this 
> technique before. Are there some docs, blog articles, descriptions and etc? 
> The only relevant documentation I have found is printf.ml :). Might someone 
> have a minimal working example to demonstrate?
>
> _______________________________________________
> Caml-list mailing list. Subscription management:
> http://yquem.inria.fr/cgi-bin/mailman/listinfo/caml-list
> Archives: http://caml.inria.fr
> Beginner's list: http://groups.yahoo.com/group/ocaml_beginners
> Bug reports: http://caml.inria.fr/bin/caml-bugs
>   
You can also take a look in format.ml, but it will be more complicate.
I don't know if it is a good idea, but if your expressions are 
complicate, try to use ocamllex and ocamlyacc.

Another thing, printf is simplistic in the syntax of the string, that is 
you only have text and no complicate expression to type (we can printf 
"x log () / 0"), but complicate in managing formatters you don't need 
(printf "it is %s o clock" time) with your examples (you have no 
parameter); from this point of view, ocamllex and ocamlyacc may be more 
relevant (and are well documented).

-- 
Cédric AUGER

Univ Paris-Sud, Laboratoire LRI, UMR 8623, F-91405, Orsay