In the current development version, the overall property score is not being computed properly
Expected: We expect this score to be the average all the individual breakdowns. Eg. the Column Shapes should be the average of all individual columns' scores
Actual: We seem to be averaging across tables first, and then averaging the averages across tables. This biases the metric, because a table with fewer columns is being counted equally compared to a table with more columns.
Steps to reproduce
from sdmetrics import load_demo
from sdmetrics.reports.multi_table import QualityReport
real_data, synthetic_data, metadata = load_demo(modality='multi_table')
report = QualityReport()
Observe that the Column Shapes property score is reported as 79.68%. However, the average of Column Shapes is actually 79.23% -- we expect this number.
Environment Details
Error Description
In the current development version, the overall property score is not being computed properly
Steps to reproduce
Observe that the Column Shapes property score is reported as 79.68%. However, the average of Column Shapes is actually 79.23% -- we expect this number.