filter

[DreamMemory001]

hello, i want to ask you some problems about filter_generated.py, how to count the number of generated molecules not in training set and in this code, i cant find how to check valency. Thank U!

[mcnaughtonadm]

Sure! So for both of these questions, there is an implementation in the original G-SchNet code but it's more specific to the QM9 datasets.
-For checking molecules between generated and training set: here
-For checking valency: here

Is this the type of thing you mean when saying you want to check valency?

In terms of our 3D-Scaffold adaptation of G-SchNet, there is no specific valency checker (see G-SchNet for why).

Just as a personal anecdote, I've never seen any duplicates between the training and generated datasets. I can definitely leave this issue open while I see if it would be simple enough to just add those functionalities back in. But for now I'd say if you really wanted to double check, to use the two implementations listed above.