This package is interesting, since it seems the only one I can find for density estimation using ME principle.

Let me ask the first question in this forum.
The current model fitting algorithms include CG, BFGS, LBFGSB, Powell, and Nelder-Mead. How about L-BFGS (unbounded optimization)? Is it implemented or is there any specific concern of not using it here?

Also, after I installed ftw, all the testing example work correctly, except bergerexamplesimulatedLBFGSB.py. I list the complete error information below (I use python 2.4.3):

$python bergerexamplesimulatedLBFGSB.py

Warning: could not load the 'pypar' parallel module.  Running in single-processor mode only.
Traceback (most recent call last):
  File "bergerexamplesimulatedLBFGSB.py", line 67, in ?
    model.fit(K, approx=True, algorithm='LBFGSB')
  File "/home/wachao/mywork/working/ftw/ftwmaxent-2.0-alpha1/ftwmaxent.py", line 665, in fit
    grad, (K,), bounds=self.bounds, pgtol=self.tol)
  File "/usr/lib/python2.4/site-packages/scipy/optimize/lbfgsb.py", line 179, in fmin_l_bfgs_b
    f[0], g = func_and_grad(x, *args)
TypeError: func_and_grad() takes exactly 1 argument (2 given)

Any comments will be highly appreciated.

Thanks!