Closed chadcwilliams closed 12 months ago
Daniel's response, which makes sense and so closing this PR:
no no the parameter condition in the generate script takes one value per condition label (which can be found in the gan-pt-file) So that means, if you trained the GAN with the binary condition conditions=cond1 then you would need to run the generate script twice Once with condition condition=0 and another time with condition=1 If you give in condition=0,1 as you suggested, the generate script would assume that the GAN was trained on two conditions (e.g. conditions=cond1,cond2) and would set cond1=0 and cond2=1 . I totally get what you are saying here with conditions=0,1 and that it would be helpful in easing the workload on the user. But the reason I've implemented it that way is simply the possibility to load more than one condition. Let's say you train the GAN on conditions=cond1,cond2 Then you would need to assign a value to these conditions in the generate script. This would be done in the that manner: conditions=0,1 Therefore the use of the conditions keyword in the generate script would not be clearly defined if the generate script would permit your proposed use of the conditions keyword. The workaround is: generate once with conditions=0 name the csv-file accordingly generate second time with conditions=1 copy & paste the content of the first csv file into the second one (or v.v.) to have all conditions in one file use this completed file for the visualization script I know that this workaround is a bit of a hassle but unfortunately I didn't had more time to build more intelligence into it..
My proposed fix for the generator. This seems to work when n_conditions=1 but I am not too sure how it should work for n_conditions>1, so maybe it's not the right solution?