example-causal-datasets
example-causal-datasets copied to clipboard
cmu-phil
"knowledge addtemporal" meaning
I'm so new to the topic, could I ask about the meaning of "knowledge addtemporal", please? I have read explaining file already, but I'm still so confused about that? Could you briefly explain it please? Thank in advance!
In case it's not clear, I mean I don't know what exactly those following rows mean "1 Region Month 2 Temperature Ws Rain RH 3 FFMC DMC DC 4 ISI BUI 5 FWI" Thank you.
Sure. The idea is that we're defining "tiers" for knowledge, where the knowledge we're asserting is that variables in later tiers can't cause variables in earlier tiers. Take these two tiers:
"1 Region Month 2 Temperature Ws Rain RH
What's being said here, for instance, is that Temperature can't cause Region, which makes sense, because you can't change which Region you're in by varying the Temperature. You could change Temperature (counterfactually) by moving to a different Region, but not vice versa.
The asterisk next to the 1 means that you can't even have had causal connections among the variable in that tier. So Region can't cause Month, and Month can't cause Region, which makes sense.
In our Tetrad application, you could load up this knowledge file and use it to guide a search, but it's being used here instead to encode what seems to be the case about the ground truth for the data. So it's interesting to see if you can make an algorithm that recovers these relationships without the crutch of knowing them in advance or can recover many of them in any case.
So, in other words, if you ran an algorithm, and it told you that Temperature causes Region, that would count against it if you take this knowledge to be correct.