inference.py

import torch
from torch.utils.data import DataLoader
import json
import os
import csv
from datetime import datetime

from nets.model import Model
from Actor.actor import Actor
from dataloader import VRP_Dataset

# --- تحميل الإعدادات ---
params_path = '/data/params_saved.json'
if not os.path.exists(params_path):
    raise FileNotFoundError(f"Settings file not found at {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 ---
model_path = "/data/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("/data", exist_ok=True)
txt_results_file = "/data/inference_results.txt"
csv_results_file = "/data/inference_results.csv"

# --- إعداد بيانات inference ---
inference_dataset = VRP_Dataset(
    size=1,
    num_nodes=params['num_nodes'],
    num_depots=params['num_depots'],
    path=dataset_path,
    device=device
)
input_size = inference_dataset.model_input_length()

# --- تحميل النموذج ---
model = Model(
    input_size=input_size,
    embedding_size=params["embedding_size"],
    decoder_input_size=params["decoder_input_size"]
)
model.load_state_dict(torch.load(model_path, map_location=device))

# --- تهيئة الممثل والـ NN Actor ---
actor = Actor(model=model,
              num_movers=params['num_movers'],
              num_neighbors_encoder=params['num_neighbors_encoder'],
              num_neighbors_action=params['num_neighbors_action'],
              device=device,
              normalize=False)
actor.eval_mode()

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 = []

for batch in dataloader:
    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")

        result_text = (
            "\n===== INFERENCE RESULT =====\n"
            f"Time: {timestamp}\n"
            f"Actor Model Total Cost: {total_time:.4f}\n"
            f"Nearest Neighbor Cost : {nn_time:.4f}\n"
            f"Improvement over NN    : {improvement:.2f}%\n"
        )
        print(result_text)
        output_lines.append(result_text)

        # حفظ النتائج 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 النتائج إلى ملف نصي ---
with open(txt_results_file, 'a') as f:
    f.write("\n".join(output_lines))
    f.write("\n=============================\n")

print(f"\n✅ Results saved to:\n- {txt_results_file}\n- {csv_results_file}")