Would be great if we could have a loop where you pass in a batch and it just iterates over it.
Here is the rough logic that I'd imagine:
- Batch gets converted to an (internal) accumulation
- Remaining gets set to the length of this accumulation
- Output "each" gives 1 item
- Optional 1: Each gives nr_batch items - so you can rebatch and run N items through a sampler at a time - this way we could even save the "Map End" construct, by emitting a arbitrarily large (nr_batch = 0) batch
- Optional 2:
Map end node starts with 1 acc_add input, if connected it spawns more inputs and outputs (would generally be cool for Loop start/end)
- Optional 3: If input is an INT, it gets expanded to list(range(int)). In addition to the "remaining" node, a float progress from 0..1 output would be nice - as well as a progress bar on the loop for the sampler
- Optional 4: CONDITIONINGs are correctly packed/unpacked/appended by this node (or acc generally) - see unpack / pack
Here is an overly complicated workflow on how i archive something like that with 6 nodes (constant number, for loop start, simplemath, accumulate, for loop close, for loop to list)
Would be great if we could have a loop where you pass in a batch and it just iterates over it.
Here is the rough logic that I'd imagine:
Map endnode starts with 1 acc_add input, if connected it spawns more inputs and outputs (would generally be cool for Loop start/end)Here is an overly complicated workflow on how i archive something like that with 6 nodes (constant number, for loop start, simplemath, accumulate, for loop close, for loop to list)
