Document value inference mechanisms

[MattiSG]

There are currently a few places where inferences take place:

• In parameters values, through the fuzzy attribute, which artificially extends the validity period of a value — Do not crash when asked for a parameter after end date #437.
• In period management, through calculate_output, which can automatically sum or divide values over periods to match the requested period.
• In period management, through set_input, which can automatically sum or divide input values over periods to match the computable period.
• In some formulas, through base_function, which can yield values that the original requested formula could not compute on its own — Make explicit input variables temporal inferences #460.
• In some formulas, through max_nb_cycles, which can block the computation toward the past and thus doesn't allow some values to be computed.

We need to agree on a common, systematic way of doing inferences, to avoid conflicting views such as in #437 or in branch extrapolate-intensive-variables-to-past.

The minimal output is a complete list of inferences and expected usage in the documentation, which could be based upon the above list.

The expected output is a set of helper functions, architectural changes and documentation that support a consistent way of inferring values in OpenFisca.

[MattiSG]

I'm being told there's also something about “suggestions” — can @cbenz detail the current behaviour, in the same format as above?

Any other known inference behaviours should be listed here. Please refrain from debating about them for the moment.

[benjello]

There is also the nb_cycles that also implictly do some time inference.

[MattiSG]

From #460 (comment):

max_nb_cycles is an implicit inference, since it may block the computation toward the past and thus doesn't allow some values to be computed.

@benjello is this the same as the nb_cycles you mentioned above?

[benjello]

Yes it is, sorry for not using the right term max_nb_cycles

[MattiSG]

Ok, it's been about 2 weeks that this reference issue was open. The time for collecting known inferences seems to be over. I updated the original post to include max_nb_cycles, and will open a PR in GitBook to document these inferences.