Skip to content

Latest commit

 

History

History
44 lines (24 loc) · 820 Bytes

File metadata and controls

44 lines (24 loc) · 820 Bytes

Pre Optimization

Ignore files without text

Testing 1001 documents across 249 folders...

Tested 1001/1001

Leave-one-out accuracy (1001 documents, 249 folders):

Top-1: 47.7% (477/1001) Top-3: 62.3% (624/1001) Top-5: 66.3% (664/1001)

Ignoring folders with only one doc

Skipping 67 documents in single-document folders.

Testing 934 documents across 182 folders...

Tested 934/934

Leave-one-out accuracy (934 documents, 182 folders):

Top-1: 51.1% (477/934) Top-3: 66.8% (624/934) Top-5: 71.1% (664/934)

Different embedding model

Skipping 65 documents in single-document folders.

Testing 947 documents across 184 folders...

Tested 947/947

Leave-one-out accuracy (947 documents, 184 folders):

Top-1: 57.3% (543/947) Top-3: 74.4% (705/947) Top-5: 81.8% (775/947)