Description : | This toolbox is a library of automatic differentiation
(http://www.autodiff.org/) software and a framework for further development of
related algorithmic computation tools.
The main tool is an abstract data type that overloads many Scilab operators and
standard functions to build the *computation graph* of a complex operation.
One can then compute information from this graph using *visitors*. This
version includes visitors for :
- Calculus of gradient, hessian, first-order jacobian and n-order directional
derivatives (using the backward mode of algorithmic derivation).
- Accurate computation of forward finite differences of a function.
- Evaluation of n-order Taylor coefficients of a univariate function.
- Re-use of a computation graph by re-evaluating the function at another point
of its domain.
See the web site for the library, as well as the included README abd INSTALL
files for more information.
|