so my first issue is, when I say after a continuation:
sol = PC.sol
and then do a
I get a
AttributeError: 'Pointset' object has no attribute 'parameterized' .
Which according to
should be an attribute of a pointset.
The 2nd thing is more general, I loosed a descent amount of time with that:
When one does a FP analysis using a user supplied Jacobian,
then PyCont seems to pick that Jacobian up and using it.
Entering a big Jacobian (for example 5 Dimensional system) is really
awkward in that string form, so there are lots of errors.
Interestingly, often one still gets the right FP classifications despite
there is a minor error in the generator.Jacobian.
But then the Conitnuation often produces just numerical
problems (like 'did not converge'). It is very intransparent for the user
to realize that it is the Jacobian which causes that trouble. The funny thing is, leaving the Jacobian
out (so initialize a generator without it) the continuations works flawlessly!!
Thanks for your note. I see there was a mistake on the website about the "parameterized" attribute. It is actually "_parameterized" and I fixed the page. There is also the "isparameterized(<pointset>)" function. This is an old part of the codebase and I'm not sure this design decision was the best, but for now I'm not going to change it.
You are right about the Jacs sometimes not helping. I've improved the Toolbox docs page to suggest using the Diff function to automatically create the Jac. While that will ensure there are no errors from hand entry, I have noticed that sometimes I cannot get a performance improvement with the Jac. After some analysis I have yet to pin down the problem but I am aware of it.
Ya I should also use the dir() function more often, I guess these _attribute is some
well thought programming convention?
I did not use the numeric Jacobian because I expected performance and rounding issues,
and I am capable of doing it by hand :P
But you would generally recommend it? Would save me some time I think..
Thanks and Greets,
Well, sometimes it's a good idea to use a leading _ but, in hindsight, I'm not certain this was one of them.
Note that the automatic Jac I'm talking about is symbolic, not numeric. Check out the Tutorial_SymbolicJac.py in PyDSTool/tests/