Use configuration file to set database collation.

[rmar3a]

Looking at how we pass around the values in the configuration file, there's probably a cleaner way to do it if we need to add/remove the list of options in the configuration file. However, this is probably good enough for now.

In addition, it looks like the collation might have been causing the non-deterministic results we were seeing in the build. Making sure the collation is consistent for all the databases (input and output) seems to make the output in the build consistent now. Guess we will have a better idea after more builds.

Update: Looks like the non-deterministic issue is still there and seems to have a different set of possible outcomes (each collation seems to have its own distinct set of possible non-deterministic outcomes) 😢

The issue is with the order of the numbers for this "table" in the BIF file that gets generated:

<DEFINITION>
	<FOR>capability(prof0,student0)</FOR>
	<GIVEN>RA(prof0,student0)</GIVEN>
	<GIVEN>salary(prof0,student0)</GIVEN>
	<TABLE> 0.166667 0.166667 0.166667 0.166667 0.166665 0.166667 0.166667 0.166667 0.166667 0.166667 0.166665 0.166667 0.166667 0.166667 0.166667 0.166667 0.166665 0.166667 0.000491 0.000491 0.000491 0.000491 0.000491 0.997545 0.008621 0.008621 0.181034 0.439655 0.353448 0.008621 0.375 0.375 0.196429 0.017857 0.017857 0.017857 0.322917 0.21875 0.427082 0.010417 0.010417 0.010417 0.166667 0.166667 0.166667 0.166667 0.166665 0.166667</TABLE>
</DEFINITION>

Not sure if the order matters, will need to look into what the values in this table are exactly (some sort of conditional probabilities?).

[oschulte]

yes they are conditional probabilities. In this case P(capability|RA, salary). That's too bad about the non-determinism. we should discuss this with Mingyi too.