jedarden/pdftract
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

fix(pdftract-60h): fix bugs in benchmark runner script

Browse source 
- Add extraction of pdftract_geomean from tool_geomeans array for regression gate
- Fix vector geomean calculation to properly pass bash array values to Python

The benchmark infrastructure was complete but had two bugs:
1. $pdftract_geomean was used but never set (line 308)
2. Vector geomean calculation had broken Python code for array expansion

These fixes ensure the regression and 10x-faster gates will work correctly
once the pdftract binary with extract/grep subcommands is available.

Refs pdftract-60h
This commit is contained in:
jedarden 2026-05-18 01:29:41 -04:00
parent 5cd0eac170
commit 398ab747fc
1 changed files with 5 additions and 2 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

7
benches/competitors/run-benchmarks.sh
View file

@ -236,6 +236,9 @@ print(math.exp(sum(math.log(v) for v in values) / len(values)))
fi
done
# Extract pdftract geomean for regression gate
local pdftract_geomean=${tool_geomeans[pdftract]:-"null"}
# Check 10x-faster gate (pdftract vs pdfminer on vector PDFs only)
# The gate applies only to vector PDFs where pdftract should excel
log_info "Computing 10x-faster gate on vector PDFs only..."
@ -270,7 +273,7 @@ print(math.exp(sum(math.log(v) for v in values) / len(values)))
if [ ${#pdftract_vector_values[@]} -gt 0 ]; then
pdftract_vector_geomean=$(python3 -c "
import math
values = ${pdftract_vector_values[@]}
values = [${pdftract_vector_values[*]}]
print(math.exp(sum(math.log(v) for v in values) / len(values)))
")
fi
@ -278,7 +281,7 @@ print(math.exp(sum(math.log(v) for v in values) / len(values)))
if [ ${#pdfminer_vector_values[@]} -gt 0 ]; then
pdfminer_vector_geomean=$(python3 -c "
import math
values = ${pdfminer_vector_values[@]}
values = [${pdfminer_vector_values[*]}]
print(math.exp(sum(math.log(v) for v in values) / len(values)))
")
fi