You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
There is some overhead to using the node structure for things, with the benefit of flexibility, breakpoints, reusing nodes in different contexts, parallelization, etc. But what if you have a workflow/macro that you're simply really happy with and want it to run as effectively as possible?
We now provide direct access to Function nodes' node_function method as a class property; in principle it should be possible to recursively "compile" macros so that instead of building nodes, they're just a function that leverages their childrens' node_function and completely hard wires how the IO gets passed between them. I.e. turn the thing into a single big python function. You would lose basically all the benefits, but gain speed from ditching the node communication.
In some cases it might be possible to go a step further and Cython the whole thing at the end to crank out even more speed.
Right now this is still a distant dream, but it's a possible goal to keep in mind.
The text was updated successfully, but these errors were encountered:
There is some overhead to using the node structure for things, with the benefit of flexibility, breakpoints, reusing nodes in different contexts, parallelization, etc. But what if you have a workflow/macro that you're simply really happy with and want it to run as effectively as possible?
We now provide direct access to
Function
nodes'node_function
method as a class property; in principle it should be possible to recursively "compile" macros so that instead of building nodes, they're just a function that leverages their childrens'node_function
and completely hard wires how the IO gets passed between them. I.e. turn the thing into a single big python function. You would lose basically all the benefits, but gain speed from ditching the node communication.In some cases it might be possible to go a step further and Cython the whole thing at the end to crank out even more speed.
Right now this is still a distant dream, but it's a possible goal to keep in mind.
The text was updated successfully, but these errors were encountered: