Spaces:
Sleeping
Sleeping
Download scripts/experiments/run_analyzer_cpu_rebaseline.bat from masterdanh/poolcoach-analyzer: direct link, hf CLI and curl.
- Browser
- Download file 2.84 kB
-
https://huggingface.co/spaces/masterdanh/poolcoach-analyzer/resolve/main/scripts/experiments/run_analyzer_cpu_rebaseline.bat
- Command line
-
hf download hf://spaces/masterdanh/poolcoach-analyzer/scripts/experiments/run_analyzer_cpu_rebaseline.bat
-
curl -L -o run_analyzer_cpu_rebaseline.bat https://huggingface.co/spaces/masterdanh/poolcoach-analyzer/resolve/main/scripts/experiments/run_analyzer_cpu_rebaseline.bat
2.84 kB
| @echo off | |
| REM ============================================================ | |
| REM CPU REBASELINE (25/08/2026) -- regenerate app/demo_analyzer/ | |
| REM from the CPU pipeline, replacing the GPU bundle of 14/08. | |
| REM | |
| REM Why this run happened (design decision, Cowork 25/08): | |
| REM The packaged demo set came from a CUDA worker (14/08, run | |
| REM c9fda71c). CUDA does NOT reproduce: 13-14/08 the same | |
| REM input/model/thresholds differed in 131 cells between two | |
| REM runs and ShotNet amplified that into a different spin | |
| REM class. The CPU path DOES reproduce bit for bit on the same | |
| REM machine (three-way check 24/08). The Space runs on CPU, so | |
| REM the shipped demo now comes from the CPU path -- and the | |
| REM older "do not re-run" rule in app/demo_analyzer/README.md, | |
| REM written for the GPU era, is replaced by "the same host MUST | |
| REM reproduce these numbers exactly". | |
| REM Second reason: the 14/08 JSON still carries Vietnamese | |
| REM strings (spin_evidence). Server strings were translated on | |
| REM 24/08, so a fresh run produces English data by itself -- | |
| REM no hand-editing of worker output. | |
| REM | |
| REM Same two inputs as run A2b (14/08) and run C (24/08): | |
| REM VOD rack_5245_5805.mp4 (Florida Open, 52:45-58:05) | |
| REM corners [[527.5,869.7],[1380.1,864.0], | |
| REM [1232.3,278.5],[673.0,272.8]] | |
| REM CPU environment: poolcoach-cv-env carries torch 2.12.1+cpu, | |
| REM so there is no CUDA to fall back to; POOLCOACH_CV_DEVICE=cpu | |
| REM states it anyway. | |
| REM | |
| REM Acceptance gate (BRIEF 25/08, step B): | |
| REM 14/14 shots must match column C of | |
| REM poolcoach-logs\results\2026-08-24_analyzer-three-way\ | |
| REM three_way.json on EVERY field (n_collisions, analytic | |
| REM v0/phi, shotnet v0/phi, the suspect valve flag). Exact | |
| REM match, no +-2 tolerance: same host, same CPU path. Any | |
| REM difference is a regression after the PR #3 merge, NOT | |
| REM noise -- stop and report, never re-run until it matches. | |
| REM | |
| REM No app or Redis has to be running: the driver starts the app | |
| REM in local mode (POOLCOACH_LOCAL_CV=1), which runs the very | |
| REM functions of scripts/cv_worker.py on a thread of the same | |
| REM process -- the same code path as the Hugging Face Space. | |
| REM ============================================================ | |
| setlocal | |
| set PYTHONUTF8=1 | |
| set PYTHONDONTWRITEBYTECODE=1 | |
| set POOLCOACH_LOCAL_CV=1 | |
| set POOLCOACH_CV_DEVICE=cpu | |
| set POOLCOACH_APP_ROOT=D:\poolcoach\poolcoach-app | |
| set POOLCOACH_ANALYZER_DIR=D:\poolcoach\out_cpu_rebaseline_20260825 | |
| set POOLCOACH_REBASELINE_VOD=D:\Khoa luan\poolcoach-rl\datasets\bb9_pilot\rack_a1\rack_5245_5805.mp4 | |
| set PY=D:\poolcoach\poolcoach-cv-env\Scripts\python.exe | |
| if not exist "%POOLCOACH_ANALYZER_DIR%" mkdir "%POOLCOACH_ANALYZER_DIR%" | |
| "%PY%" "D:\poolcoach\run_analyzer_cpu_rebaseline.py" | |
| echo [rebaseline] exit code %ERRORLEVEL% | |
| endlocal | |