Tweaks to demo - error I don't understand

[mohawk2]

These changes are hopefully self-explanatory and slightly improve the code. But on running the demo (in order to add it to the website, as discussed), it's giving:

OOPS!!! Something went wrong, please make a bug report!: lu_backsub: LU decomposition is undef -- probably from a singular matrix.
 at lib/PDL/MatrixOps.pd line 1175.
	PDL::MatrixOps::lu_backsub(undef, undef, undef, PDL=SCALAR(0xb90677060)) called at (path)/PDL/Fit/LM.pm line 124
	PDL::lmfit(PDL=SCALAR(0xb907a3498), PDL=SCALAR(0xb907b2060), PDL=SCALAR(0xb9079ebb8), CODE(0xb8f14a5b8), PDL=SCALAR(0xb90674dc8), HASH(0xb9079e990)) called at (path)/blib/lib/Physics/Ellipsometry/VASE.pm line 173
	Physics::Ellipsometry::VASE::fit(Physics::Ellipsometry::VASE=HASH(0xb907a9c00), PDL=SCALAR(0xb90674dc8)) called at (eval 187) line 31
	eval 'package main; no strict; use PDL; 
  # Fit the multi-layer model to the real 3-angle VASE data
  $ml_vase->set_model(\\&multilayer_model);
[...]

The error is in this bit:

      print "  LM from grid best did not converge, trying refined guess...\n";
      $ml_fit = $ml_vase->fit(pdl [2.12, 1.8, 0.1, $best_d, 1.0, 0.50]);

Before throwing the exception, it reported:

---- Output:
Grid search: best thickness = 230 nm (d_s = 2.30)
Fitting with LM refinement...
  LM from grid best did not converge, trying refined guess...

It may be necessary to add more error-catching in PDL::Fit::LM so it's caught a bit more clearly, and/or prints out the data it had an error with. Let me know if so.

[mohawk2]

A further issue comes from the point I mentioned in our emails: I believe it would be a good idea (and I hope pretty easy) to make a PDL::IO::Ellipsometry module in this distro. The immediate problem I'm having is, when I try to regenerate the website, it transforms the demo text into this:

DOING (package PDL::Demos::Ellipsometry; *do_output=\&main::do_output; sub do_output; no strict; use PDL;
  ($fh, $tmpfile) = tempfile(SUFFIX => '.dat', UNLINK => 1);
do_output + $fh "# Wavelength(nm) Angle(deg) Psi(deg) Delta(deg)\n";
  for my $i (0 .. $wavelength->nelem - 1) {
do_output sprintf $fh "%.1f  %.1f  %.4f  %.4f\n",
          $wavelength->at($i), $angle->at($i),
          $psi->at($i), $delta->at($i);
  }
  close $fh;

  $vase = Physics::Ellipsometry::VASE->new(layers => 1);
  $data = $vase->load_data($tmpfile);

Because the not-very-clever heuristic wants to change print into a call to the output-capture function, do_output. If you were writing to your tempfile with a completely separate function not defined in the demo, it wouldn't have a problem. The problem it is getting here is the file-handle, which is only valid for real print, and in any case I wouldn't want to capture that here.

Separately, is there another way to get data loaded without going via a file? That feels like a further way to improve this situation.

[jtrujil43]

Yes for the demo I should put the sample data in the code itself. Will do that in the next iteration.