CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

Project Purpose

Poker analysis tool for GGPoker using a Logitech StreamCam pointed at a 2560×1440 monitor. Captures the game via webcam, reads cards via OCR and color detection, calculates equity via Monte Carlo simulation.

Running the Scripts

# Live camera viewer with settings panel (trackbars)
python3 main.py

# Poker analyzer — board OCR + Monte Carlo equity
python3 analyzer.py              # auto OCR hero cards
python3 analyzer.py Qd Qc        # manually specify hero cards (rank+suit)

# Webcam scraper using dickreuter_poker backend
python3 webcam_scraper.py

Architecture

main.py            — Camera viewer with OpenCV trackbars. Loads/saves settings.json.
analyzer.py        — Main analysis pipeline: webcam → crop → OCR → Monte Carlo → advice.
webcam_scraper.py  — Adapter connecting webcam to dickreuter_poker TableScraper.
virtualbox/        — Stub module: dickreuter_poker imports virtualbox which isn't needed.
files/             — Card sprite sheet (table_card.png + card.json) for future template matching.
settings.json      — Persisted camera settings (auto-tuned, do not edit manually).
tmp/               — All temporary files, snapshots, debug images go here.

External dependency: dickreuter_poker

Located at /home/isaevea/python/dickreuter_poker. Added to sys.path at runtime.

poker.decisionmaker.montecarlo_numpy2.Evaluation — Monte Carlo equity calculator (CPU/numpy, no GPU).
poker.scraper.table_scraper.TableScraper — Table state scraper (card templates from https://dickreuter.com:7778/).
poker.tools.mongo_manager.MongoManager — Fetches GGPoker table config from remote API. Table name: "Official GGPoker 6player".

GUI modules of dickreuter_poker (PyQt6) are stubbed out in webcam_scraper.py via sys.modules injection.

Camera Hardware

Device: Logitech StreamCam → /dev/video2 (index 2). /dev/video0 is UVC Camera.
Optimal settings (auto-tuned, saved in settings.json): focus=60 (manual), sharpness=192, exposure=500 (manual).
Qt font issue fix: Liberation Sans fonts copied to .venv/lib/python3.12/site-packages/cv2/qt/fonts/.
Window titles must be ASCII-only — Cyrillic in cv2.namedWindow() causes NULL pointer crash. Trackbar labels support Cyrillic.

Webcam Coordinate System

All coordinate constants in analyzer.py are in the 1280×720 webcam frame:

BOARD_AREA  = (220, 370, 330, 640)  # y1,y2,x1,x2 — entire flop/turn/river area
HERO_L_AREA = (438, 478, 472, 530)  # left hole card
HERO_R_AREA = (438, 478, 530, 585)  # right hole card

These are camera-position-dependent. If camera moves, recalibrate by running a snapshot and checking tmp/ images.

In webcam_scraper.py, the GGPoker window boundary:

TABLE_X1, TABLE_Y1 = 15, 55
TABLE_X2, TABLE_Y2 = 760, 685

This crops the GGPoker client from the webcam frame and scales to 1500×1100 (dickreuter's CROP_WIDTH × CROP_HEIGHT).

Card Detection

Board cards (read_board_cards): BOARD_AREA is divided into 3 equal columns. Top-left corner of each is OCR'd via Tesseract (tessedit_char_whitelist=23456789TJQKA). Suit detected by mean red channel (>100 → hearts/diamonds, else spades).

Hero cards (read_hero_cards): Color-based suit detection in HSV space — blue hue → ♦, green → ♣, red → ♥, dark → ♠. GGPoker uses emoji-style colored card backgrounds.

Tessdata path: /home/isaevea/python/dickreuter_poker/tessdata/

Monte Carlo Card Format

montecarlo_numpy2.Evaluation.run_evaluation(card1, card2, tablecards, iterations, player_amount)

Cards encoded as [rank, suit] where rank: 2–14 (A=14), suit: 0=clubs, 1=diamonds, 2=hearts, 3=spades.

RANK_MAP = {'2':2,...,'T':10,'J':11,'Q':12,'K':13,'A':14}
SUIT_MAP = {'c':0,'d':1,'h':2,'s':3}

Returns float equity (0.0–1.0). Typical: >0.65 → raise, >0.45 → call, <0.25 → fold.

4.0 KiB Raw Permalink Blame History Unescape Escape