And not only by dataframe. We wind up very often jumping through hoops to unpack the results. I think the dataframe is exactly the response we want from the .iter and .zip methods, but when manually constructing a for_node, it seems like having separate output channels full of list-like data for each iterated output will be more useful. One can always recreate a dataframe from this with the InputToDataframe transformer node, but I also think we don't need to throw the dataframe representation away -- for_node is a function defining a new class, so we are free to modify output channels at that stage and can have a output_dataframe: bool = False option.
@JNmpi @ligerzero-ai any thoughts? Han, I'll ping you for review when I finally get a PR for this.
And not only by dataframe. We wind up very often jumping through hoops to unpack the results. I think the dataframe is exactly the response we want from the
.iterand.zipmethods, but when manually constructing afor_node, it seems like having separate output channels full of list-like data for each iterated output will be more useful. One can always recreate a dataframe from this with theInputToDataframetransformer node, but I also think we don't need to throw the dataframe representation away --for_nodeis a function defining a new class, so we are free to modify output channels at that stage and can have aoutput_dataframe: bool = Falseoption.@JNmpi @ligerzero-ai any thoughts? Han, I'll ping you for review when I finally get a PR for this.