At the moment, the blaster steps corresponding to edges in the blaster graph are run from root nodes to leaf nodes (unless the output nodes for a given step already exist). A better approach would be to run this backwards: start from the leaf nodes, check if they exist, and if not then run the step(s) needed them. Repeat if the intermediate blaster files for those nodes need to be computed as well.
One benefit of this is that it naturally solves #16, since the step needed to get rec.crg.pdb would not need to be run since it is provided by the user, thereby eliminating rec.pdb as a required input.
It should be noted that this feature has already been coded up and is working on the beam-search branch.
ianscottknight
commented
1 year ago
At the moment, the blaster steps corresponding to edges in the blaster graph are run from root nodes to leaf nodes (unless the output nodes for a given step already exist). A better approach would be to run this backwards: start from the leaf nodes, check if they exist, and if not then run the step(s) needed them. Repeat if the intermediate blaster files for those nodes need to be computed as well.
One benefit of this is that it naturally solves #16, since the step needed to get rec.crg.pdb would not need to be run since it is provided by the user, thereby eliminating rec.pdb as a required input.
It should be noted that this feature has already been coded up and is working on the beam-search branch.