Computational Data Flow Expression / DAG Builder about xtensor HOT 8 CLOSED

You can already reuse an expression and evaluate it when needed, however named variables are not supported:

xt::array<double> x, y;
auto exp_res1 = x + y + 3; // exp_res1 type is xfunction<....>, not evaluated
auto exp_res2 = exp_res1 / 2; // exp_res2 type is xfunction<...>, not evaluated

// init x and y
x = arr1;
y = arr2;
// Forces evaluation of exp_res1 in res1. exp_res1 remains unchanged.
xt::xarray<double> res1 = exp_res1;

// changes y
y = arr3;
// Forces evaluation of exp_res1 (with new y) in res2. exp_res1 remains unchanged.
xt::xarray<double> res2 = exp_res1;

// Forces evaluation of exp_res1 (with new y) in res2. exp_res1 remains unchanged.
xt::xarray<double> res3 = exp_res2;

from xtensor.

What happens when x, y is not in scope?

from xtensor.

xfunction doesn't provide any way to retrieve its arguments, so you have to store references to x and y somewhere, that you can reach from the code that modifies them. I guess it can be problematic if you have sophisticated expressions with dozens of variables.

@SylvainCorlay any thought on implementing named variables ?

from xtensor.

Also something like

xt::xarray<double> res = xt::xfunction("x + y + 3").setp("x", arr1).setp("y", arr2).eval()

res = xt::xfunction("f(x, y) = x + y + 3").setp("x", arr1).setp("y", arr2).eval()
res = xt::xfunction("f(x, y) = x + y + 3").setp(0, arr1).setp(1, arr2).eval()

res = xt::xfunction("f(x, y) = g(x) + h(y) + 3").setf("g", f1).setf("h", f2).setp("x", arr1).setp("y", arr2).eval()

Also have some predefined standard functions

xt::sin, xt::ln, xt::reg, .....

from xtensor.

You're not supposed to call xfunction constructor by yourself.
So this code:

xt::array<double> res = xt:xfunction("x + y + 3")

should definitely be replaced by:

xt::array<double> res = x + y + 3;

or

auto exp_res = x + y + 3;

if you want to keep the expression for a deferred evaluation.

Besides, building xfunction based on string expression is not supported. And won't. Never. This library is a computation library, not a parser library. If you need such a feature, feel free to implement it as an aside project, but xtensor is not the right place where to do it.

Almost all the standard mathematical functions defined in the C++ standard library have their counterpart in xtensor. Check the xmath.hpp header and the doc. If you need one that is missing, please open another issue dedicated to this, so we don't mix up unrelated features.

from xtensor.

As mentioned before, what if I do not have x and y when I set up the expression.

from xtensor.

That's the same problem as the first one. The solution is to provide a way to retrieve the arguments of an xfunction (either with indexed or named getters), not a constructor that parses a string.

from xtensor.

Thanks for this @sirinath I am closing this since it is out of scope at the moment.

from xtensor.