feat: app.py

app.py

@@ -6,94 +6,95 @@ from io import BytesIO
 from PIL import Image
 from typing import Dict, List, Any
 
-# 导入项目的核心逻辑和配置
+# Import core logic and configurations from the project
 from geo_bot import (
     GeoBot,
     AGENT_PROMPT_TEMPLATE,
     BENCHMARK_PROMPT,
-)  # 导入Prompt模板以供复用
+)
 from benchmark import MapGuesserBenchmark
 from config import MODELS_CONFIG, DATA_PATHS, SUCCESS_THRESHOLD_KM
 from langchain_openai import ChatOpenAI
 from langchain_anthropic import ChatAnthropic
 from langchain_google_genai import ChatGoogleGenerativeAI
 
-# --- 页面UI设置 ---
+# --- Page UI Setup ---
 st.set_page_config(page_title="MapCrunch AI Agent", layout="wide")
 st.title("🗺️ MapCrunch AI Agent")
-st.caption("一个通过多步交互探索和识别地理位置的AI智能体")
+st.caption(
+    "An AI agent that explores and identifies geographic locations through multi-step interaction."
+)
 
-# --- Sidebar用于配置 ---
+# --- Sidebar for Configuration ---
 with st.sidebar:
-    st.header("⚙️ 运行配置")
+    st.header("⚙️ Agent Configuration")
 
-    # 从HF Secrets获取API密钥 (部署到HF Spaces时，需要在Settings->Secrets中设置)
+    # Get API keys from HF Secrets (must be set in Space settings when deploying)
     os.environ["OPENAI_API_KEY"] = st.secrets.get("OPENAI_API_KEY", "")
     os.environ["ANTHROPIC_API_KEY"] = st.secrets.get("ANTHROPIC_API_KEY", "")
-    # 添加其他你可能需要的API密钥
     # os.environ['GOOGLE_API_KEY'] = st.secrets.get("GOOGLE_API_KEY", "")
 
-    model_choice = st.selectbox("选择AI模型", list(MODELS_CONFIG.keys()))
+    model_choice = st.selectbox("Select AI Model", list(MODELS_CONFIG.keys()))
     steps_per_sample = st.slider(
-        "每轮最大探索步数", min_value=3, max_value=20, value=10
+        "Max Exploration Steps per Sample", min_value=3, max_value=20, value=10
     )
 
-    # 加载golden labels以供选择
+    # Load golden labels for selection
     try:
         with open(DATA_PATHS["golden_labels"], "r", encoding="utf-8") as f:
             golden_labels = json.load(f).get("samples", [])
         total_samples = len(golden_labels)
         num_samples_to_run = st.slider(
-            "选择测试样本数量", min_value=1, max_value=total_samples, value=3
+            "Number of Samples to Test", min_value=1, max_value=total_samples, value=3
         )
     except FileNotFoundError:
-        st.error(f"数据文件 '{DATA_PATHS['golden_labels']}' 未找到。请先准备数据。")
+        st.error(
+            f"Data file '{DATA_PATHS['golden_labels']}' not found. Please prepare the data."
+        )
         golden_labels = []
         num_samples_to_run = 0
 
     start_button = st.button(
-        "🚀 启动Agent Benchmark", disabled=(num_samples_to_run == 0), type="primary"
+        "🚀 Start Agent Benchmark", disabled=(num_samples_to_run == 0), type="primary"
     )
 
-# --- Agent运行逻辑 ---
+# --- Agent Execution Logic ---
 if start_button:
-    # 准备运行环境
+    # Prepare the environment
     test_samples = golden_labels[:num_samples_to_run]
 
     config = MODELS_CONFIG.get(model_choice)
     model_class = globals()[config["class"]]
     model_instance_name = config["model_name"]
 
-    # 初始化用于统计结果的辅助类和列表
+    # Initialize helpers and result lists
     benchmark_helper = MapGuesserBenchmark()
     all_results = []
 
     st.info(
-        f"即将开始Agent Benchmark... 模型: {model_choice}, 步数: {steps_per_sample}, 样本数: {num_samples_to_run}"
+        f"Starting Agent Benchmark... Model: {model_choice}, Steps: {steps_per_sample}, Samples: {num_samples_to_run}"
     )
 
-    # 创建一个总进度条
-    overall_progress_bar = st.progress(0, text="总进度")
+    overall_progress_bar = st.progress(0, text="Overall Progress")
 
-    # 初始化Bot (注意：在HF Spaces上，必须以headless模式运行)
-    # 将Bot的初始化放在循环外，可以复用同一个浏览器实例，提高效率
-    with st.spinner("正在初始化浏览器和AI模型..."):
+    # Initialize the bot outside the loop to reuse the browser instance for efficiency
+    with st.spinner("Initializing browser and AI model..."):
+        # Note: Must run in headless mode on HF Spaces
         bot = GeoBot(model=model_class, model_name=model_instance_name, headless=True)
 
-    # 主循环，遍历所有选择的测试样本
+    # Main loop to iterate through all selected test samples
     for i, sample in enumerate(test_samples):
         sample_id = sample.get("id", "N/A")
         st.divider()
-        st.header(f"▶️ 运行中:样本 {i + 1}/{num_samples_to_run} (ID: {sample_id})")
+        st.header(f"▶️ Running Sample {i + 1}/{num_samples_to_run} (ID: {sample_id})")
 
-        # 加载地图位置
         if not bot.controller.load_location_from_data(sample):
-            st.error(f"加载样本 {sample_id} 失败，已跳过。")
+            st.error(f"Failed to load location for sample {sample_id}. Skipping.")
             continue
 
         bot.controller.setup_clean_environment()
 
-        # 为当前样本创建可视化布局
+        # Create the visualization layout for the current sample
         col1, col2 = st.columns([2, 3])
         with col1:
             image_placeholder = st.empty()
@@ -101,25 +102,25 @@ if start_button:
             reasoning_placeholder = st.empty()
             action_placeholder = st.empty()
 
-        # --- 内部的Agent探索循环 ---
+        # --- Inner agent exploration loop ---
         history = []
         final_guess = None
 
         for step in range(steps_per_sample):
             step_num = step + 1
             reasoning_placeholder.info(
-                f"思考中... (第 {step_num}/{steps_per_sample} 步)"
+                f"Thinking... (Step {step_num}/{steps_per_sample})"
             )
             action_placeholder.empty()
 
-            # 观察并标记箭头
+            # Observe and label arrows
             bot.controller.label_arrows_on_screen()
             screenshot_bytes = bot.controller.take_street_view_screenshot()
             image_placeholder.image(
                 screenshot_bytes, caption=f"Step {step_num} View", use_column_width=True
             )
 
-            # 更新历史
+            # Update history
             history.append(
                 {
                     "image_b64": bot.pil_to_base64(
@@ -129,7 +130,7 @@ if start_button:
                 }
             )
 
-            # 思考
+            # Think
             prompt = AGENT_PROMPT_TEMPLATE.format(
                 remaining_steps=steps_per_sample - step,
                 history_text="\n".join(
@@ -157,12 +158,12 @@ if start_button:
             )
             action_placeholder.success(f"**AI Action:** `{action}`")
 
-            # 强制在最后一步进行GUESS
+            # Force a GUESS on the last step
             if step_num == steps_per_sample and action != "GUESS":
-                st.warning("已达最大步数，强制执行GUESS。")
+                st.warning("Max steps reached. Forcing a GUESS action.")
                 action = "GUESS"
 
-            # 行动
+            # Act
             if action == "GUESS":
                 lat, lon = (
                     decision.get("action_details", {}).get("lat"),
@@ -171,10 +172,10 @@ if start_button:
                 if lat is not None and lon is not None:
                     final_guess = (lat, lon)
                 else:
-                    # 如果AI没在GUESS时提供坐标，再问一次
-                    # (这里的简化处理是直接结束，但在更复杂的版本可以再调用一次AI)
-                    st.error("GUESS动作中缺少坐标，本次猜测失败。")
-                break  # 结束当前样本的探索
+                    st.error(
+                        "GUESS action was missing coordinates. Guess failed for this sample."
+                    )
+                break  # End exploration for the current sample
 
             elif action == "MOVE_FORWARD":
                 bot.controller.move("forward")
@@ -185,9 +186,9 @@ if start_button:
             elif action == "PAN_RIGHT":
                 bot.controller.pan_view("right")
 
-            time.sleep(1)  # 在步骤之间稍作停顿，改善视觉效果
+            time.sleep(1)  # A brief pause between steps for better visualization
 
-        # --- 单个样本运行结束，计算并展示结果 ---
+        # --- End of single sample run, calculate and display results ---
         true_coords = {"lat": sample.get("lat"), "lng": sample.get("lng")}
         distance_km = None
         is_success = False
@@ -197,23 +198,23 @@ if start_button:
             if distance_km is not None:
                 is_success = distance_km <= SUCCESS_THRESHOLD_KM
 
-            st.subheader("🎯 本轮结果")
+            st.subheader("🎯 Round Result")
             res_col1, res_col2, res_col3 = st.columns(3)
             res_col1.metric(
-                "最终猜测 (Lat, Lon)", f"{final_guess[0]:.3f}, {final_guess[1]:.3f}"
+                "Final Guess (Lat, Lon)", f"{final_guess[0]:.3f}, {final_guess[1]:.3f}"
             )
             res_col2.metric(
-                "真实位置 (Lat, Lon)",
+                "Ground Truth (Lat, Lon)",
                 f"{true_coords['lat']:.3f}, {true_coords['lng']:.3f}",
             )
             res_col3.metric(
-                "距离误差",
+                "Distance Error",
                 f"{distance_km:.1f} km" if distance_km is not None else "N/A",
-                delta=f"{'成功' if is_success else '失败'}",
+                delta=f"{'Success' if is_success else 'Failure'}",
                 delta_color=("inverse" if is_success else "off"),
             )
         else:
-            st.error("Agent 未能做出最终猜测。")
+            st.error("Agent failed to make a final guess.")
 
         all_results.append(
             {
@@ -226,22 +227,27 @@ if start_button:
             }
         )
 
-        # 更新总进度条
+        # Update overall progress bar
         overall_progress_bar.progress(
-            (i + 1) / num_samples_to_run, text=f"总进度: {i + 1}/{num_samples_to_run}"
+            (i + 1) / num_samples_to_run,
+            text=f"Overall Progress: {i + 1}/{num_samples_to_run}",
         )
 
-    # --- 所有样本运行完毕，显示最终摘要 ---
-    bot.close()  # 关闭浏览器
+    # --- End of all samples, display final summary ---
+    bot.close()  # Close the browser
     st.divider()
-    st.header("🏁 Benchmark 最终摘要")
+    st.header("🏁 Benchmark Summary")
 
     summary = benchmark_helper.generate_summary(all_results)
     if summary and model_choice in summary:
         stats = summary[model_choice]
         sum_col1, sum_col2 = st.columns(2)
-        sum_col1.metric("总成功率", f"{stats.get('success_rate', 0) * 100:.1f} %")
-        sum_col2.metric("平均距离误差", f"{stats.get('average_distance_km', 0):.1f} km")
-        st.dataframe(all_results)  # 显示详细结果表格
+        sum_col1.metric(
+            "Overall Success Rate", f"{stats.get('success_rate', 0) * 100:.1f} %"
+        )
+        sum_col2.metric(
+            "Average Distance Error", f"{stats.get('average_distance_km', 0):.1f} km"
+        )
+        st.dataframe(all_results)  # Display the detailed results table
     else:
-        st.warning("没有足够的结果来生成摘要。")
+        st.warning("Not enough results to generate a summary.")