This project expresses scientific computation in a form more like symbolic mathematics. It performs derivatives on that representation, and then converts to a regular programming language. It's similar to auto-differentiation, but the differentiation is performed on a different representation. It can also be thought of as mixed auto and symbolic differentiation.

The representation is a structured collection of statements with the symbolic part using Sympy.

The steps directory contains the development progression.

The most complete versions of the code and tests are in the derivcodegen and tests directories.

A function is constructed as

    x = Symbol("x")
    e1 = Symbol("e1")
    R = Routine("test1_d1x")
    R.inputs = [x]
    R.outputs = [e1]
    R.stmts = [Statement(e1, 2 * x ** 2 + 3 * x)]

or there is a more compact string input, intended for testing

    text1 = """
        e0 = 2*x**2 + 3*x
    """

    R = routine_from_text("test1", text1)

Take the derivative

    x = Symbol('x')
    dR = R.diff(x)

Generate the code to a function (Julia in this example. Python and C++ are also supported targets). The resulting function is

function test1_d1x(x)
    e0 = 2 * x .^ 2 + 3 * x
    e0_d1x = 4 * x + 3
    return (e0, e0_d1x)
end