Implementation Questions

[J-Gann]

Dear authors,

Thank you for making the implementation of your work publicly available. After reading your paper SymRAG some questions arised. I would appreciate your help on this:

Regarding the dynamic threshold adjustment, I am asking myself why you reduce T_high_k given a low performing neural component. Would this update not in fact increase the neural band and lead to more usage of the neural path instead of less usage?

symrag/src/system/system_logic_helpers.py

Line 106 in 1f6a3ea

updated_thresholds['high_complexity_thr'] -= 0.05

Additionally I am asking myself why for the symbolic path to be chosen, low resource pressure is necessary. I assumed the purpose of symbolic execution is to reduce computational load and hence should be used in high resource pressure scenarios.

symrag/src/system/system_logic_helpers.py

Line 31 in 1f6a3ea

if query_complexity < low_complexity_thr and overall_resource_pressure < low_resource_thr:

I would really appreciate your feedback on these points to help my understanding.

Best regards,
Jonas Gann

--
PhD Student @ Data Science Research Group
Institute of Computer Science (IfI)

Heidelberg University
https://ds.ifi.uni-heidelberg.de/