After the recent addition of the AFLOW adapter to propnet, it became apparent that the current algorithm for Graph, while generally robust, will not be feasible to execute when conditions (temperature, etc.) and condition-dependent properties/models are added to the graph.
If you add an elastic tensor, a structure, a band gap, and a temperature condition to a Material on propnet and ask Graph to derive properties, you will eventually end up with an algorithm cycle where over 700 million possible input sets are generated. Even parallelized, this is a pretty daunting number that will only get larger as more properties and more conditions are added.
As touched on in #216, there needs to be some pipeline we use to aggregate or throw out quantities based on uncertainty. This will reduce the number of quantities available to create input sets and keep it down to a manageable number.
Fielding comments and suggestions here or offline.
After the recent addition of the AFLOW adapter to propnet, it became apparent that the current algorithm for Graph, while generally robust, will not be feasible to execute when conditions (temperature, etc.) and condition-dependent properties/models are added to the graph.
If you add an elastic tensor, a structure, a band gap, and a temperature condition to a Material on propnet and ask Graph to derive properties, you will eventually end up with an algorithm cycle where over 700 million possible input sets are generated. Even parallelized, this is a pretty daunting number that will only get larger as more properties and more conditions are added.
As touched on in #216, there needs to be some pipeline we use to aggregate or throw out quantities based on uncertainty. This will reduce the number of quantities available to create input sets and keep it down to a manageable number.
Fielding comments and suggestions here or offline.