FAQs
- Q: During clustering or result merging, why does Google Protobuf report the error "exceeded maximum protobuf size of 2gb" and the size of the result is 0, when the SaveDataToFile API is called to serialize and save the result to my local PC through Google Protobuf?
A: FeatureClustering uses the third-party Google Protobuf to perform serialization. For Protobuf, the size of the serialized object must be no more than 2 GB. Therefore, if the clustering result is larger than 2 GB, use the SaveData() API to obtain the clustering result from the memory.
- Q: During clustering, the chip is lost. What is the possible cause?
A: Evaluate the data volume before clustering. Currently, the theoretical memory of a single chip is 8 GB, and the theoretical memory for storing input data is 6 GB. If the size of the input base library for nearest neighbor (NN) calculation is greater than 6 GB, unpredictable consequences may occur. It is recommended that more chips be used to perform clustering in parallel.