Solve: Error in PROGN [or a callee]: Bind stack overflow.


On Tue, 23 Nov 2010, Julian Andres Klode wrote:

< On Di, 2010-11-23 at 15:29 +0000, Leo Butler wrote:
< > 
< > On Tue, 23 Nov 2010, Julian Andres Klode wrote:
< > 
< > < When trying to run the code below, I get 
< > < 	Error in PROGN [or a callee]: Bind stack overflow.
< > < 
< > < The function preg() is doing regression, returning a term approximating
< > < the points P. The term is based on 'term', with the variables in 'vars'
< > < being replaced. When running on a term a^x, solve() results in a bind
< > < stack overflow; for a * x it works perfectly.
< > < 
< > < Hope you can help, as I need this function to work.
< > < 
< > < -- Maxima version information:
< > < Maxima version: 5.22.1
< > < Maxima build date: 1:16 11/6/2010
< > < Host type: x86_64-unknown-linux-gnu
< > < Lisp implementation type: GNU Common Lisp (GCL)
< > < Lisp implementation version: GCL 2.6.7
< > < 
< > < -- Code:
< > < preg(P, term, vars) := block([eqns, g, res, ratprint: false],
< > <     g : sum((P[i][2] - subst(P[i][1], x, term))^2, i, 1, length(P)),
< > <     eqns : makelist(diff(g,vars[i]), i, 1, length(vars)),
< > <     subst(solve(eqns, vars)[1], term)
< > < )$
< > < 
< > < P : [[1650, 0.545], [1750, 0.728], [1800, 0.906],
< > <      [1850, 1.171], [1900, 1.608], [1950, 2.517]]$
< > < 
< > < preg(P, a ^ x * b, [a, b]);
< >  
< >  Is there a reason you are trying to find an exact solution, when
< >  your input data are floats? The lbfgs package provides an iterative
< >  floating-point solver. The lsquares package also provides an interface
< >  that you can use without needing to write lots of code. Both are
< >  documented in the manual.
< >  Leo
< I just implemented a straight-forward solution to the problem. I did not
< find the packages, probably because I only searched for solve, and
< completely ignored the possibility to look for square-related stuff.
< 
< I also noticed that the solution I am looking for is a^x*b, not a^x+b. I
< then used lsquares_estimates() and everything works, although the
< performance seems to be very slow, at least compared to what I recall to
< have seen from MuPad users.
< 
< In case you are interested, my final function is:
<         preg2(P, term, vars) := block([p, vals, res],
<             p: apply(matrix, P),
<             vals: lsquares_estimates(p, ['x,'y], 'y=term, vars),
<             res : subst(vals[1], term),
<             if (pregplot) then
<                 wxplot2d([res, ['discrete, P]], ['x, P[1][1]-5, P[length(P)][1]+5], 
<                          ['legend, "N?herungsfunktion",  "Punkte"],
<                          ['style, 'lines, 'points]),
<             float(res)
<         );
< 
< 
< Which takes a list of points (P), a term with x (term), and a list of
< variables (vars). It supports printing of the points and the created
< function if pregplot is true. It's just a bit more simplified interface
< for my use cases now (which are: find a function and plot it).
 
 Your problem can be recast as a loglinear model (the objective function
 is different, but I doubt this matters for you):

 Try:

 preg2(log(P),log(a^x*b),[a,b]), logexpand=all;

 The logexpand=all effects the expansion to log(a)*x+log(b).
 The model is linear in log(a) and log(b), and the solution is
 instantaneous.

Leo
-- 
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.