Performance

[lrennels]

There should be places we can really improve performance, but ideally we would maintain the flexibility for countries and time, while also speeding things up. Some steps might be:

• remove use of unique
• remove use of Query looping functions like @filter
• run Profiler for type stability

For example, it may be faster to move away from DataFrames and Query to use lower-level indexing methods like replacing

 subset = g_ssp_datasets[ssp_dict_key] |>
            @filter(_.year == gettime(t) && _.country in p.country_names) |>
            DataFrame

with something like (pseudocode)

yr_idxs = g_ssp_datasets[ssp_dict_key].year == gettime(t) # Bit Vector
country_idxs = indexin(g_ssp_datasets[ssp_dict_key]. country,  p.country_names) # Vector of Ints

and then use those to pare down to the subset, or are those also slow?