I think it is okay to check for variance in fixed features only in predictive strategies. This helps to add data from only one task to the random strategy (or any strategy without surrogate) in case you have a multi-task domain. This might be useful in case you have a stepwise strategy with a random step to get initial random data from one of the tasks where you might not have historical data of.
I think it is okay to check for variance in fixed features only in predictive strategies. This helps to add data from only one task to the random strategy (or any strategy without surrogate) in case you have a multi-task domain. This might be useful in case you have a stepwise strategy with a random step to get initial random data from one of the tasks where you might not have historical data of.