Looking for TVM contributions

[robve]

(06-18-2024 06:15 PM)Albert Chan Wrote:  I was looking at your test samples pdf. #23 just quit, returned NaN

That's because guess_i() returns i%=101.94 that npmt cannot handle. If I go back to a small initial I% it works fine. I just left it in for now, knowing that iguess() should be changed later after investigating before deciding. By the way, this iguess() function is your suggested initial best guess.

It is the same test as on the contributions list of benchmarks #6.

- Rob

EDIT: I realize it may be a bit difficult to see the TVM parameters in the PDF. They are n, i% (to solve with ?), pv, pmt, fv below (the last one is a repeat which I neglected to take out):

./tvm 40 ? 900 -40 -1000 1
./tvm 40 ? 900 1000 -1000 1
./tvm 40 ? 900 -400 -1000 1
./tvm 31536000 ? 0 -0.01 331667.0067
./tvm 525600 ? 0 -0.01 5282.36776893622391
./tvm 525600 ? 0 -0.01 5260.38242600032663
./tvm 32 ? -999999 0 1000000
./tvm 360 ? 0 -1000 1000000
./tvm 328 ? 35000 -324 0
./tvm 58 ? -775 -49.56 4000
./tvm 32 ? -6000 0 10000
./tvm 48 ? 19198.60 -450 0
./tvm 5 ? -369494.09 17500 540000
./tvm 348 ? 243400 -1363.29 0
./tvm 30 ? -3200 -717.44 60000
./tvm 60 ? 243400 -1363.29 -222975.98
./tvm 24 ? 0 -50.26 1281.34
./tvm 6 ? -32000 0 28346.96
./tvm 60 ? 8000 -150.97 0
./tvm 456 ? 270000 -1215.3333333333 0
./tvm 480 ? 100000 -208.333333333333343 0
./tvm 10 ? 50 -30 400
./tvm 10 ? 50 -30 80
./tvm 10 ? -100 10 1e-10
./tvm 32 ? -999999 0 1e6
./tvm 525600 ? 0 -0.01 5260.38242600032663
./tvm 40 ? 900 -400 -1000 1

EDIT 2: the nan problem with iguess() returning a rate larger than 100% can be fixed with a test such as:

Code:

  i = iguess();
  if (i == 0.0)
    i = 1e-5;
  else if (fabs(i) >= 100)
    i = copysign(50, i);

Then we get an answer:
evals=7, i%=-36.8933698741777363%, FVerror=1.4e-14

But is +/-50% a reasonable new starting point to always avoid this? Don't think so. It looks like your idea to use edge is nice.