Python Guide & Performance Benchmark - XAD Automatic Differentiation (original) (raw)

The following example for first-order adjoint mode illustrates how to use it:

[](#%5F%5Fcodelineno-0-1)import xad.adj_1st as xadj [](#%5F%5Fcodelineno-0-2) [](#%5F%5Fcodelineno-0-3) [](#%5F%5Fcodelineno-0-4)# set independent variables [](#%5F%5Fcodelineno-0-5)x0_ad = xadj.Real(1.0) [](#%5F%5Fcodelineno-0-6)x1_ad = xadj.Real(1.5) [](#%5F%5Fcodelineno-0-7)x2_ad = xadj.Real(1.3) [](#%5F%5Fcodelineno-0-8)x3_ad = xadj.Real(1.2) [](#%5F%5Fcodelineno-0-9) [](#%5F%5Fcodelineno-0-10)with xadj.Tape() as tape: [](#%5F%5Fcodelineno-0-11) # and register them [](#%5F%5Fcodelineno-0-12) tape.registerInput(x0_ad) [](#%5F%5Fcodelineno-0-13) tape.registerInput(x1_ad) [](#%5F%5Fcodelineno-0-14) tape.registerInput(x2_ad) [](#%5F%5Fcodelineno-0-15) tape.registerInput(x3_ad) [](#%5F%5Fcodelineno-0-16) [](#%5F%5Fcodelineno-0-17) # start recording derivatives [](#%5F%5Fcodelineno-0-18) tape.newRecording() [](#%5F%5Fcodelineno-0-19) [](#%5F%5Fcodelineno-0-20) # calculate the output [](#%5F%5Fcodelineno-0-21) y = x0_ad + x1_ad - x2_ad * x3_ad [](#%5F%5Fcodelineno-0-22) [](#%5F%5Fcodelineno-0-23) # register and seed adjoint of output [](#%5F%5Fcodelineno-0-24) tape.registerOutput(y) [](#%5F%5Fcodelineno-0-25) y.derivative = 1.0 [](#%5F%5Fcodelineno-0-26) [](#%5F%5Fcodelineno-0-27) # compute all other adjoints [](#%5F%5Fcodelineno-0-28) tape.computeAdjoints() [](#%5F%5Fcodelineno-0-29) [](#%5F%5Fcodelineno-0-30) # output results [](#%5F%5Fcodelineno-0-31) print(f"y = {y}") [](#%5F%5Fcodelineno-0-32) print(f"first order derivatives:\n") [](#%5F%5Fcodelineno-0-33) print(f"dy/dx0 = {x0_ad.derivative}") [](#%5F%5Fcodelineno-0-34) print(f"dy/dx1 = {x1_ad.derivative}") [](#%5F%5Fcodelineno-0-35) print(f"dy/dx2 = {x2_ad.derivative}") [](#%5F%5Fcodelineno-0-36) print(f"dy/dx3 = {x3_ad.derivative}")

The Python bindings follow largely the same syntax and workflow as in C++.

Modules and Naming

Module	Description	Contents
xad	The main module, which contain global functions and subpackages	value, derivative
xad.exceptions	Contains all exceptions, with the same names as described in Exceptions	e.g. NoTapeException
xad.math	Mirrors Python's math module, with functions for XAD's active types.	e.g. sin, exp
xad.fwd_1st	Active type for first-order forward mode	Real
xad.adj_1st	Active type for first-order adjoint mode as well as the corresponding tape type	Real, Tape

Notes

First order forward mode (module xad.fwd_1st) and first order adjoint mode are supported (module xad.adj_1st)
The active type is called Real in all modes
In adjoint mode, a newly constructed Tape object is not automatically activated on construction. It can be activated using tape.activate() later, but we recommend using a with block as illustrated in the example above.
The math functions in xad.math have been designed as a drop-in replacement for the standard Python math module. They not only support calls with XAD's active type, but also with regular float variables.
Checkpointing and external function features are not yet supported in Python.
The x.getDerivative() and x.setDerivative() methods of active types are also available as the Python property x.derivative with both set and get functionality.
The x.getValue() method of active types is also available as the read-only property x.value
Use y.setDerivative(1.0) or the property setter y.derivative = 1.0 to seed and access derivatives.
Complex numbers are not yet supported in the Python bindings.