Load only a subset of input columns at the start of each period about liam2

[gdementen]

1. you cannot currently do that: it loads all columns present in the input dataset if there are any rows for the current period in the input dataset. So you can have values for some periods but not others and some persons but not others but you need to have all input columns for each person that is present for a particular period. This is a stupid restriction but it is all due to the missing value problem (for int we cannot distinguish a value of -1 from a missing value/column for an individual, and it's even worse for booleans). When that problem will be fixed, extending the merge function (merge_arrays in data.py) to check for missing values should be relatively easy.

2. yes, values loaded at the start of a period simply replace the value from the previous period, but are not treated differently during the simulation, so they are modified. If you want to keep the input value if present, you have to explicitly test it with something like (assuming a float column):

# column == column is meant to filter out nan (missing) values
column_in_input: column == column and column != lag(column)
column: if(column_in_input, column, expr_to_compute_column)

[AlexisEidelman]

Thank you.

About the second point, I switched to an other tactics. I created an entity futur which looks like my entity person. With a link 'in_futur' from person to futur id i can write:

column: if(in_futur.id > -1, in_futur.column, expr_to_compute_column)