Dear Chris, Your fix gives correct results for my equations. Since esub is called recursively to code tends to become pretty slow for larger equations (esub goes at least 25 levels deep). I am looking forward to your work on cse.
With kind regards, Pieter Eendebak On Apr 20, 7:26 pm, "Chris Smith" <[email protected]> wrote: > Pieter wrote: > >> Dear newsgroup, > > >> I am trying to generate common sub expressions from an equation, but > >> the results are incorrect for negative values of the symbols. > >> An example (sympy 0.6.7) > > >> x1=Symbol('x1') > >> x2=Symbol('x2') > >> y1=Symbol('y1') > >> z=Symbol('z', negative=True) > > >> eq=x1*(1+ y1*(x1**2/z**2+x2**2/z**2))/z > >> v1,v2=cse(eq) > >> print(eq) > >> print(v1) > >> print(v2) > > >> For x1=1, x2=1, y1=0 and z=-1 the original equation and the value of > >> thecseexpression yield different values. > > >> Is this a bug incse, or can I make certain assumptions on the > >> symbols to generate correct expressions? > > This is happening because the substitutions of the repeated pattern (z**2) > are targetting the z. This is a known issue that I am working on; my > subs_cleanup branch is a first step toward this. Here's a nuisance-level fix > (not robust for every type of sympy object but should work for plain > algebraic expressions): > > def esub(eq, reps): > '''make exact replacements in eq using the old/new key/value > pairs of reps''' > if eq in reps: > return reps[eq] > args = list(eq.args) > if not args: > return eq > for i, a in enumerate(args): > args[i] = esub(a, reps) > return eq.func(*args) > > >>> esub(eq, dict([(v, k) for k, v in v1])) # whatcseshould give > x1*(1 + y1*(x0*x1**2 + x0*x2**2))/z > >>> eq.subs({x1:1,x2:1,y1:0,z:-1}) # original > -1 > >>> esub(eq, dict([(v, k) for k, v in v1])).subs({x1:1,x2:1,y1:0,z:-1}) # > esub-cse > -1 > > /chris -- You received this message because you are subscribed to the Google Groups "sympy" group. To post to this group, send email to [email protected]. To unsubscribe from this group, send email to [email protected]. For more options, visit this group at http://groups.google.com/group/sympy?hl=en.
