Duplicate Results

[yosr-nabli]

1. Entity Filters
   The paper identifies date/time tokens as the most common redundant entities. The correct per-dataset filters are:
   ​

HDFS — no filter needed. Block IDs (blk_XXXXXX) and IPs are genuine entities and should pass through unchanged.

BGL / TDB — filter the compound timestamp-correlation token (e.g. 1117838570-2005-06-03-15.42.50.675872) that appears in every log line and would connect all logs together as false positives

But for example for HDFS i don't get the same # of Entity nodes as reported in the paper. Can you provide the Filtering for HDFS, TDB and BGL.

2. Non-Grouped Sampling — Missing Implementation
   This is never implemented in the code. The dataset loaders always attach a structured group ID to each log — block_id for HDFS, node_id for BGL, user_id/component_id for TDB — meaning every sample is inherently grouped by one of those fields.
   Can you provide the implementation for this as well