I am using guild to manage my runs of machine learning Hyperparameter tuning, I have a group of runs that differ by only two of tens flags while the other ~10 flags have the same value (typical when you want to see the influence of a subset of hyper parameters to the result), I would like to know if there is a way automatically detect the columns that have various value and only print those columns plus the scalar columns.
For example, if I have flags A~Z, and scalar output error, in one experiment I would like to see the influence of A and Z only on error, I run a batch runs with different A and Z, and would like to know how to print the column A and Z only without look at 24 columns.
I can do it by `guild compare -cc ‘=A,=Z,error’, but I want an automatic way without manually pick and input the name of columns that are different among runs.
Guild doesn’t seem to support this function unless you tab into the code, I would recommend writing an alias to do that if you tend to view only several columns frequently.