grad evaluation - Githubissues

dutitello / paransys2

A Python module that allows you to run ANSYS APDL scripts setting selected parameters values by Python, and returning the values of all parameters in the end of the analysis.

https://dutitello.github.io/paransys2/

MIT License

1 stars 0 forks source link

grad evaluation #3

Open dutitello opened 4 years ago

dutitello commented 4 years ago

It will be determined by finite difference method. Forward, central and backward methods at first.

All scalar parameters should be evaluated for each declarated parameter:

If the parameters from the APDL are b, h, l, stress and it's passed to grad just the value of b we could determine db/db (why would we do that?), dh/db, dl/db and dstress/db. If b,h,l are input parameters the derivatives of that will be 0, but all could be calculated and the user will define what it's useful for it (dstress/db).
It's possible to determine it for arrays/tables??

dutitello commented 4 years ago

It's partially done!

No support for arrays parameters in paransys yet;
The three methods works;
It's possible to choose in relation to what grad is evaluated with (notfor=[] and onlyfor=[])
It returns a diactionary in the form grad('y','x') = y'(x) = dy/dx.

dutitello commented 4 years ago

Arrays and tables are for now out. grad looks ok!

dutitello commented 4 years ago

the name of the function is now derivatives! Since grad(f(X)) are the derivatives of f(X) for all X variables the output should be an dictionary for each output parameter (f) and each dictionary should have it's derivatives.

dutitello commented 4 years ago

I used for central diference method h/2 steps... but it's a good ideia for numerical evaluations... It need to be always h...