Use of presamples in LCA and index advancement could be improved

[PascalLesage]

Suppose a presamples package containing matrix data with the following properties:

• with ncols>1, say 5
• seed='sequential'
• presamples package dirpath= pp_dirpath.

If used in MonteCarloLCA, the indices start at 0 (thanks to the reset of the indexers here). For each iteration (i.e. each time next(mc_object) is called), the matrix data indexer goes up by 1. All this is sane and easy to use.

For LCA, however, things are more complicated.

Suppose the data in the presamples package is some time series, and we want to generate LCA results for each time step by calling LCA n times. If we just use the LCA.lci() method, things don't work:

>>> import brightway2 as bw
>>> import presamples as ps
>>> lca = bw.LCA({some_act:1}, presamples=[pp_dirpath])
>>> print("iteration\tindex")
>>> for i in range(5):
...     lca.lci()
...     print(i, "\t", lca.presamples.matrix_indexer[0].index)

iteration   index
0        0
1        0
2        0
3        0
4        0

We need to actively update indices to get the indices moving up. If we naively use update_matrices, we don't get the first column.

>>> lca = bw.LCA({some_act:1}, presamples=[pp_dirpath])
>>> print("iteration\tindex")
>>> for i in range(5):
...     lca.presamples.update_matrices()
...     lca.lci()
...     print(i, "\t", lca.presamples.matrix_indexer[0].index)

iteration   index
0        1
1        2
2        3
3        4
4        5

Resetting sequential indices, like for MonteCarloLCA, gives the expected behaviour to the problem described above:

>>> lca = bw.LCA({some_act:1}, presamples=[pp_dirpath])
>>> lca.presamples.reset_sequential_indices()
>>> print("iteration\tindex")
>>> for i in range(5):
...     lca.presamples.update_matrices()
...     lca.lci()
...     print(i, "\t", lca.presamples.matrix_indexer[0].index)

iteration   index
0        0
1        1
2        2
3        3
4        4

This resetting could simply be added here Tried it, no presamples tests fail.

We may want to consider moving indices directly when the lci method is invoked.

[PascalLesage]

The simple fix above has tests failing in bw.LCA.

[PascalLesage]

Tests in LCA show that the expected behaviour is not what would be allowed by the change described above.

Indeed, the expected behaviour is that:

• an LCA object whose presamples have not had their matrices updated via lca.presamples.update_matrices will use the values in the first column (this is also how things work with the change above). This is good, because LCA will often use presamples with a single column (scenario)
• the first call of update_matrices moves the index from 0 to 1. This is because the  indexer.count is 1 when loaded. In the suggestion above, the  indexer.count is reset to 0, so the first call of update_matrices actually moves the  indexer.count to 1 but the index itself remains at 0.

So the expected usage is sequentially seeded presample packages in LCA would be something like this:

>>> lca = bw.LCA({some_act:1}, presamples=[pp_dirpath])
>>> print("iteration\tindex")
>>> for i in range(5):
...     if i != 0:
...         lca.presamples.update_matrices()
...     lca.lci()
...     print(i, "\t", lca.presamples.matrix_indexer[0].index)

iteration   index
0        0
1        1
2        2
3        3
4        4

The if clause is a bit annoying, so we need to decide what approach is better:

• Weirdness because the index remains at 0 after calling update_indices for the first time (suggestion above); or
• Only calling update_indices on runs beyond the first.

@cmutel any preference?