Update inference.py

inference.py

@@ -1,9 +1,10 @@
 import torch
-from torch.utils.data import DataLoader
+from torch.utils.data import DataLoader, TensorDataset
 import json
 import os
 import csv
 from datetime import datetime
+import pandas as pd
 import numpy as np
 
 from nets.model import Model
@@ -12,6 +13,7 @@ from dataloader import VRP_Dataset
 
 # --- تحديد المجلد الآمن ---
 safe_data_dir = "/home/user/data"
+orders_file = os.path.join(safe_data_dir, "orders.csv")
 
 # --- تحميل الإعدادات ---
 params_path = os.path.join(safe_data_dir, 'params_saved.json')
@@ -21,41 +23,46 @@ if not os.path.exists(params_path):
 with open(params_path, 'r') as f:
     params = json.load(f)
 
-# --- تعيين الجهاز ---
 device = params['device']
-dataset_path = params['dataset_path']
 
-# --- مسار النموذج المحفوظ ---
+# --- تحميل النموذج ---
 model_path = os.path.join(safe_data_dir, "model_state_dict.pt")
 if not os.path.exists(model_path):
     raise FileNotFoundError(f"Model not found at {model_path}")
 
-# --- تحضير مجلد النتائج ---
-os.makedirs(safe_data_dir, exist_ok=True)
-txt_results_file = os.path.join(safe_data_dir, "inference_results.txt")
-csv_results_file = os.path.join(safe_data_dir, "inference_results.csv")
-
-# --- إعداد بيانات inference ---
-inference_dataset = VRP_Dataset(
-    dataset_size=1,
-    num_nodes=params['num_nodes'],
-    num_depots=params['num_depots'],
-    dataset_path=dataset_path,
-    device=device
-)
-
-input_size = inference_dataset.model_input_length()
-
-# --- تحميل النموذج ---
 model = Model(
-    input_size=input_size,
+    input_size=params["num_nodes"] * 4,  # تعديل حسب تمثيل البيانات
     embedding_size=params["embedding_size"],
     decoder_input_size=params["decoder_input_size"]
 )
 model.load_state_dict(torch.load(model_path, map_location=device))
 model.eval()
 
-# --- تهيئة الممثل والـ NN Actor ---
+# --- إعداد ملف النتائج ---
+txt_results_file = os.path.join(safe_data_dir, "inference_results.txt")
+csv_results_file = os.path.join(safe_data_dir, "inference_results.csv")
+os.makedirs(safe_data_dir, exist_ok=True)
+
+# --- قراءة ملف CSV ---
+if not os.path.exists(orders_file):
+    raise FileNotFoundError(f"orders.csv not found at {orders_file}")
+
+df = pd.read_csv(orders_file)
+
+# --- تجهيز البيانات كـ Tensor ---
+pickup_lng = df['lng'].to_numpy()
+pickup_lat = df['lat'].to_numpy()
+delivery_lng = df['delivery_gps_lng'].to_numpy()
+delivery_lat = df['delivery_gps_lat'].to_numpy()
+
+# دمج البيانات في مصفوفة واحدة (batch_size=1, N, features)
+coords = np.stack([pickup_lng, pickup_lat, delivery_lng, delivery_lat], axis=1)  # shape (N, 4)
+coords_tensor = torch.tensor(coords, dtype=torch.float32).unsqueeze(0).to(device)
+
+# --- إنشاء Batch مدخل للنموذج ---
+batch = [{"coords": coords_tensor}]
+
+# --- إعداد الممثلين ---
 actor = Actor(model=model,
               num_movers=params['num_movers'],
               num_neighbors_encoder=params['num_neighbors_encoder'],
@@ -66,58 +73,59 @@ actor = Actor(model=model,
 nn_actor = Actor(model=None, num_movers=1, num_neighbors_action=1, device=device)
 nn_actor.nearest_neighbors()
 
-# --- استدلال وتخزين النتائج ---
-dataloader = DataLoader(inference_dataset, batch_size=1, collate_fn=inference_dataset.collate)
-output_lines = []
-summary_text = ""
-input_summary = ""
-
-for batch in dataloader:
-    with torch.no_grad():
-        data = batch[0]  # ← تعديل هام
-        actor.greedy_search()
-        actor_output = actor(batch)
-        total_time = actor_output['total_time'].item()
-
-        nn_output = nn_actor(batch)
-        nn_time = nn_output['total_time'].item()
-
-        improvement = (nn_time - total_time) / nn_time * 100
-        timestamp = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
-
-
-        # --- نص كامل للعرض والطباعة ---
-        result_text = (
-            "\n===== INFERENCE RESULT =====\n"
-            f"Time: {timestamp}\n"
-            f"Actor Model Total Cost: {total_time:.4f} units\n"
-            f"Nearest Neighbor Cost : {nn_time:.4f} units\n"
-            f"Improvement over NN    : {improvement:.2f}%\n"
-        )
-        print(result_text)
-        output_lines.append(result_text)
-
-        # --- للواجهة Gradio ---
-        summary_text = (
-            f"🕒 Time: {timestamp}\n"
-            f"🚚 Actor Cost: {total_time:.4f} units\n"
-            f"📍 NN Cost: {nn_time:.4f} units\n"
-            f"📈 Improvement: {improvement:.2f}%\n\n"
-            f"{input_summary}"
-        )
-
-        # --- CSV حفظ ---
-        write_header = not os.path.exists(csv_results_file)
-        with open(csv_results_file, 'a', newline='') as csvfile:
-            writer = csv.writer(csvfile)
-            if write_header:
-                writer.writerow(["Timestamp", "Actor Cost", "NN Cost", "Improvement (%)"])
-            writer.writerow([timestamp, f"{total_time:.4f}", f"{nn_time:.4f}", f"{improvement:.2f}"])
-
-# --- حفظ النتائج إلى ملف نصي ---
-with open(txt_results_file, 'a') as f:
-    f.write("\n".join(output_lines))
-    f.write("\n=============================\n")
-
-# --- طباعة للـ UI
-print(f"\n🔍 Summary for UI:\n{summary_text}")
+# --- تشغيل النموذج ---
+with torch.no_grad():
+    actor.greedy_search()
+    actor_output = actor(batch)
+    total_time = actor_output['total_time'].item()
+
+    nn_output = nn_actor(batch)
+    nn_time = nn_output['total_time'].item()
+
+    improvement = (nn_time - total_time) / nn_time * 100
+    timestamp = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
+
+    # --- عرض جزء من البيانات المستخدمة ---
+    coords_preview = "\n".join([
+        f"Order {i}: P=({lng1:.4f},{lat1:.4f}) → D=({lng2:.4f},{lat2:.4f})"
+        for i, (lng1, lat1, lng2, lat2) in enumerate(coords[:5])
+    ])
+    if coords.shape[0] > 5:
+        coords_preview += f"\n... (showing 5 of {coords.shape[0]} orders)"
+
+    input_summary = f"📌 Input Orders Preview:\n{coords_preview}"
+
+    # --- نتيجة كاملة للطباعة ---
+    result_text = (
+        "\n===== INFERENCE RESULT =====\n"
+        f"Time: {timestamp}\n"
+        f"Actor Model Total Cost: {total_time:.4f} units\n"
+        f"Nearest Neighbor Cost : {nn_time:.4f} units\n"
+        f"Improvement over NN    : {improvement:.2f}%\n"
+    )
+    print(result_text)
+
+    # --- واجهة Gradio ---
+    summary_text = (
+        f"🕒 Time: {timestamp}\n"
+        f"🚚 Actor Cost: {total_time:.4f} units\n"
+        f"📍 NN Cost: {nn_time:.4f} units\n"
+        f"📈 Improvement: {improvement:.2f}%\n\n"
+        f"{input_summary}"
+    )
+
+    # --- حفظ إلى CSV ---
+    write_header = not os.path.exists(csv_results_file)
+    with open(csv_results_file, 'a', newline='') as csvfile:
+        writer = csv.writer(csvfile)
+        if write_header:
+            writer.writerow(["Timestamp", "Actor Cost", "NN Cost", "Improvement (%)"])
+        writer.writerow([timestamp, f"{total_time:.4f}", f"{nn_time:.4f}", f"{improvement:.2f}"])
+
+    # --- حفظ نصي ---
+    with open(txt_results_file, 'a') as f:
+        f.write(result_text)
+        f.write("\n=============================\n")
+
+    # --- إخراج للواجهة
+    print(f"\n🔍 Summary for UI:\n{summary_text}")