memory leakage when inference about chemprop

[swansonk14]

Hi @HANSANGJUN,

Thank you for the question! There should be some degree of RAM usage increase during inference since we are creating a list that holds the results across all batches of the test set on which inference is being performed. However, the memory usage should not be very large. What sort of RAM usage numbers are you seeing? Also just to clarify, is this during inference on the validation or test set in train.py or is this when running predict.py?

Thanks,
Kyle

[sjhan91]

I modified your code to combined with my algorithms, and I found caching mechanism
if cache: SMILES_TO_GRAPH[d.smiles] = mol_graph
filled my RAM.

I disallowed that mechanism. Thank you.

[swansonk14]

Okay great, I'm glad you found that!