.gitignore

@@ -7,3 +7,4 @@ results
 .claude
 node_modules
 src/paddle_ocr/wheels
+src/*.log

src/run_raytune.py

@@ -0,0 +1,124 @@
+#!/usr/bin/env python3
+"""Ray Tune hyperparameter search for PaddleOCR via REST API."""
+
+import os
+from datetime import datetime
+
+import requests
+import pandas as pd
+
+import ray
+from ray import tune, train, air
+from ray.tune.search.optuna import OptunaSearch
+
+# Configuration
+WORKER_PORTS = [8001, 8002]
+OUTPUT_FOLDER = "results"
+os.makedirs(OUTPUT_FOLDER, exist_ok=True)
+
+# Search space
+search_space = {
+    "use_doc_orientation_classify": tune.choice([True, False]),
+    "use_doc_unwarping": tune.choice([True, False]),
+    "textline_orientation": tune.choice([True, False]),
+    "text_det_thresh": tune.uniform(0.0, 0.7),
+    "text_det_box_thresh": tune.uniform(0.0, 0.7),
+    "text_det_unclip_ratio": tune.choice([0.0]),
+    "text_rec_score_thresh": tune.uniform(0.0, 0.7),
+}
+
+
+def trainable_paddle_ocr(config):
+    """Call PaddleOCR REST API with the given hyperparameter config."""
+    import requests
+    from ray import train
+
+    NUM_WORKERS = len(WORKER_PORTS)
+
+    context = train.get_context()
+    trial_id = context.get_trial_id() if context else "0"
+    try:
+        trial_num = int(trial_id.split("_")[-1])
+    except (ValueError, IndexError):
+        trial_num = hash(trial_id)
+
+    worker_idx = trial_num % NUM_WORKERS
+    api_url = f"http://localhost:{WORKER_PORTS[worker_idx]}"
+
+    payload = {
+        "pdf_folder": "/app/dataset",
+        "use_doc_orientation_classify": config.get("use_doc_orientation_classify", False),
+        "use_doc_unwarping": config.get("use_doc_unwarping", False),
+        "textline_orientation": config.get("textline_orientation", True),
+        "text_det_thresh": config.get("text_det_thresh", 0.0),
+        "text_det_box_thresh": config.get("text_det_box_thresh", 0.0),
+        "text_det_unclip_ratio": config.get("text_det_unclip_ratio", 1.5),
+        "text_rec_score_thresh": config.get("text_rec_score_thresh", 0.0),
+        "start_page": 5,
+        "end_page": 10,
+    }
+
+    try:
+        response = requests.post(f"{api_url}/evaluate", json=payload, timeout=None)
+        response.raise_for_status()
+        metrics = response.json()
+        metrics["worker"] = api_url
+        train.report(metrics)
+    except Exception as e:
+        train.report({
+            "CER": 1.0,
+            "WER": 1.0,
+            "TIME": 0.0,
+            "PAGES": 0,
+            "TIME_PER_PAGE": 0,
+            "worker": api_url,
+            "ERROR": str(e)[:500]
+        })
+
+
+def main():
+    # Check workers
+    print("Checking workers...")
+    for port in WORKER_PORTS:
+        try:
+            r = requests.get(f"http://localhost:{port}/health", timeout=10)
+            print(f"  Worker {port}: {r.json().get('status', 'unknown')}")
+        except Exception as e:
+            print(f"  Worker {port}: ERROR - {e}")
+
+    print("\nStarting Ray Tune...")
+    ray.init(ignore_reinit_error=True)
+
+    tuner = tune.Tuner(
+        trainable_paddle_ocr,
+        tune_config=tune.TuneConfig(
+            metric="CER",
+            mode="min",
+            search_alg=OptunaSearch(),
+            num_samples=64,
+            max_concurrent_trials=len(WORKER_PORTS),
+        ),
+        run_config=air.RunConfig(verbose=2, log_to_file=True),
+        param_space=search_space,
+    )
+
+    results = tuner.fit()
+
+    # Save results
+    df = results.get_dataframe()
+    timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
+    filepath = os.path.join(OUTPUT_FOLDER, f"raytune_paddle_results_{timestamp}.csv")
+    df.to_csv(filepath, index=False)
+    print(f"\nResults saved: {filepath}")
+
+    # Best config
+    if len(df) > 0 and "CER" in df.columns:
+        best = df.loc[df["CER"].idxmin()]
+        print(f"\nBest CER: {best['CER']:.6f}")
+        print(f"Best WER: {best['WER']:.6f}")
+
+    ray.shutdown()
+
+
+if __name__ == "__main__":
+    main()