143 requests more dimension for NIL review. With the new proposed dimensions it is unclear how the dimensions should be weighted. The first dry run results show that for example the sandbox dimensions can result to a very higher Z score (8.8 for "1", -0.1 for "0")

How is this problem misaligned with goals of app mining? It should be clear for app miners how the dimensions are used in the NIL score. The different dimensions will have not equal weights as they depend on the performance of the other apps.

What is the explicit recommendation you’re looking to propose? Use the sum of all 6 dimensions and then calculate the Z and theta score.

What is the dry run period (if any) No dry run needed as the current dry run data can be updated.

Describe your long term considerations in proposing this change. Please include the ways you can predict this recommendation could go wrong and possible ways mitigate. The NIL score can be used better by 3rd party app listings to compare a subset of apps.

Additional context

143

friedger commented 4 years ago

//cc @larrysalibra Also related to #217

larrysalibra commented 4 years ago

The PBC team decides how to take the numbers we give them to generate a score. Happy to have this changed if everyone thinks it makes sense! cc @hstove

stacks-archive / app-mining

Weight of NIL dimensions #209

143 requests more dimension for NIL review. With the new proposed dimensions it is unclear how the dimensions should be weighted. The first dry run results show that for example the sandbox dimensions can result to a very higher Z score (8.8 for "1", -0.1 for "0")

143