Try and find a quicker alternative to `recurse_tagged_database`

[pjamesjoyce]

presamples makes recalculating brightway LCA objects with different methods and parameter sets lightning fast.

The full analysis calculation still drags because of the need to explore the whole graph for each method for each parameter set e.g. from recent project:

• 18 methods, 6 parameter sets = 108 calculations
• time to calculate LCA/LCI/LCIA x 108 with presamples = 1.5 seconds
• time to run lcopt analysis = 2 min 40 seconds

Maybe look into exploring the whole foreground graph once then recharacterising the tagged groups per method? (groups don't change between iterations)

[pjamesjoyce]

Benchmark analysis time down to 2 minutes with new analysis setup - it's a start - 25% faster...

[pjamesjoyce]

Slightly new approach to the analysis step

Benchmark analysis is down to 1 min 8 seconds - more than twice as fast as before.

Details in brief

• new functions editing bw2analyzer.traverse_tagged_databases, bw2analyzer.aggregate_tagged_graph and bw2analyzer.recurse_tagged_database to allow multiple methods to be used in one run, and to allow secondary tags to be applied to graph nodes

  The graph returned is the same except the impact key in technosphere/biosphere becomes a list with n elements (where n in the number of methods used). When aggregated, the value in the dictionary is also now a list with one sum per method.

• new function multi_traverse_tagged_databases only needs to run once (for all methods) instead of multiple times (once per method). This saves a bit of time (~25% faster).

• addition of secondary tags means that the tree can be tagged by name (for aggregation of foreground results) and by lcopt_type (for tree/bullseye/mass flow visualisations) in one run, instead of having to run tagging by type, then run tagging by name. This saves a lot of time as the analysis only needs to run once, not twice.

@cmutel: If you think that a version of the tagged analysis that can use multiple methods and has secondary tags would be useful for bw2analyzer proper I'm happy to write them up as a pull request (with the proper tests etc.)

[cmutel]

@pjamesjoyce Yes, please do!