Update app.py

app.py

@@ -1,85 +1,80 @@
-# lstm_cnn_app.py (modified)
-"""
-Robust Gradio app for CNN-LSTM fault classification.
-
-Features added:
-- Prefer local model file; optionally download from Hugging Face Hub if HUB_REPO/HUB_FILENAME set.
-- If no model found, app still starts but prediction functions return friendly message.
-- Port selection:
-    * If GRADIO_SERVER_PORT or PORT env var is set, try that.
-    * Otherwise find a free ephemeral port and use it.
-    * If binding fails, fall back to demo.launch() with no explicit port (Gradio picks).
-- Reduces TF logging noise via TF_CPP_MIN_LOG_LEVEL (optional).
-"""
+# app.py -- Spaces-ready robust version for lstm_cnn model
 import os
-import socket
+import threading
+import traceback
 import numpy as np
 import pandas as pd
 import gradio as gr
-from tensorflow.keras.models import load_model
 from huggingface_hub import hf_hub_download
 
-# Reduce TensorFlow log noise (keeps warnings but hides info/debug)
-os.environ.setdefault("TF_CPP_MIN_LOG_LEVEL", "2")
+# Use Keras model loader (change if you have PyTorch)
+from tensorflow.keras.models import load_model
+
+# ---------------- Config ----------------
+LOCAL_MODEL_FILE = os.environ.get("LOCAL_MODEL_FILE", "lstm_cnn_model.h5")
+HUB_REPO = os.environ.get("HUB_REPO", "")       # optional: "username/repo"
+HUB_FILENAME = os.environ.get("HUB_FILENAME", "")  # optional: "lstm_cnn_model.h5"
+# ----------------------------------------
 
-# CONFIG: change these if your model filename/repo are different
-LOCAL_MODEL_FILE = "lstm_cnn_model.h5"
-HUB_REPO = ""         # e.g., "username/lstm-cnn-model" (leave empty to disable)
-HUB_FILENAME = ""     # e.g., "lstm_cnn_model.h5"
+MODEL = None
+MODEL_READY = False
+MODEL_LOAD_ERROR = None
 
 def download_from_hub(repo: str, filename: str):
     try:
-        print(f"Downloading {filename} from {repo} ...")
+        print(f"[model] Downloading {filename} from {repo} ...", flush=True)
         path = hf_hub_download(repo_id=repo, filename=filename)
-        print("Downloaded to:", path)
+        print("[model] Downloaded to:", path, flush=True)
         return path
     except Exception as e:
-        print("Failed to download from hub:", e)
+        print("[model] Hub download failed:", e, flush=True)
         return None
 
-def get_model_path():
-    # Prefer local file
-    if os.path.exists(LOCAL_MODEL_FILE):
-        return LOCAL_MODEL_FILE
-    # Try env override for local path (handy in Spaces)
-    alt = os.environ.get("MODEL_FILE_PATH")
-    if alt and os.path.exists(alt):
-        return alt
-    # Try hub
-    if HUB_REPO and HUB_FILENAME:
-        return download_from_hub(HUB_REPO, HUB_FILENAME)
-    return None
-
-def try_load_model(path):
+def load_model_background():
+    global MODEL, MODEL_READY, MODEL_LOAD_ERROR
     try:
-        m = load_model(path)
-        print("Loaded model:", path)
-        return m
-    except Exception as e:
-        print("Failed to load model:", e)
-        return None
-
-MODEL_PATH = get_model_path()
-MODEL = try_load_model(MODEL_PATH) if MODEL_PATH else None
-
+        model_path = None
+        if os.path.exists(LOCAL_MODEL_FILE):
+            model_path = LOCAL_MODEL_FILE
+            print(f"[model] Found local model: {model_path}", flush=True)
+        elif HUB_REPO and HUB_FILENAME:
+            model_path = download_from_hub(HUB_REPO, HUB_FILENAME)
+        else:
+            print("[model] No local model file and no HUB_REPO/HUB_FILENAME configured.", flush=True)
+
+        if model_path is None:
+            raise FileNotFoundError("Model file not found locally or on Hugging Face Hub.")
+
+        print(f"[model] Loading model from {model_path} ...", flush=True)
+        MODEL = load_model(model_path)
+        MODEL_READY = True
+        print("[model] Model loaded OK.", flush=True)
+    except Exception:
+        MODEL_LOAD_ERROR = traceback.format_exc()
+        MODEL_READY = False
+        print("[model] Error loading model:\\n", MODEL_LOAD_ERROR, flush=True)
+
+# Start model loader in background so Gradio can bind to PORT immediately
+loader = threading.Thread(target=load_model_background, daemon=True)
+loader.start()
+
+# ---------------- Helper functions ----------------
 def prepare_input_array(arr, n_timesteps=1, n_features=None):
     arr = np.array(arr)
     if arr.ndim == 1:
         if n_features is None:
-            # assume arr is flattened timesteps*features
             return arr.reshape(1, n_timesteps, -1)
         return arr.reshape(1, n_timesteps, int(n_features))
     elif arr.ndim == 2:
-        # treat as (timesteps, features) -> add batch dim
-        if arr.shape[0] == 1:
-            return arr.reshape(1, arr.shape[1], -1)
         return arr
     else:
         return arr
 
 def predict_text(text, n_timesteps=1, n_features=None):
-    if MODEL is None:
-        return "模型未加载。请上传 'lstm_cnn_model.h5' 到 Space 根目录，或设置 HUB_REPO/HUB_FILENAME。"
+    if not MODEL_READY:
+        if MODEL_LOAD_ERROR:
+            return f"模型加载失败：\\n{MODEL_LOAD_ERROR}"
+        return "模型尚���加载完成，请稍候（后台正在加载）。"
     try:
         arr = np.fromstring(text, sep=',')
         x = prepare_input_array(arr, n_timesteps=int(n_timesteps), n_features=(int(n_features) if n_features else None))
@@ -90,8 +85,10 @@ def predict_text(text, n_timesteps=1, n_features=None):
         return f"预测失败: {e}"
 
 def predict_csv(file, n_timesteps=1, n_features=None):
-    if MODEL is None:
-        return {"error": "模型未加载。请上传 'lstm_cnn_model.h5' 到 Space 根目录，或设置 HUB_REPO/HUB_FILENAME。"}
+    if not MODEL_READY:
+        if MODEL_LOAD_ERROR:
+            return {"error": f"模型加载失败：\\n{MODEL_LOAD_ERROR}"}
+        return {"error": "模型尚未加载完成，请稍候（后台正在加载）。"}
     try:
         df = pd.read_csv(file.name)
         X = df.values
@@ -103,13 +100,16 @@ def predict_csv(file, n_timesteps=1, n_features=None):
     except Exception as e:
         return {"error": f"预测失败: {e}"}
 
-# Gradio UI
+# ---------------- Gradio UI ----------------
 with gr.Blocks() as demo:
-    gr.Markdown("# CNN-LSTM Fault Classification")
-    if MODEL is None:
-        gr.Markdown("**注意**：未检测到模型 (lstm_cnn_model.h5)。请上传模型或在代码中设置 HUB_REPO/HUB_FILENAME。应用仍会启动，但预测不可用。")
+    gr.Markdown("# CNN-LSTM Fault Classification (Spaces)")
+    if MODEL_READY:
+        gr.Markdown("模型已加载 ✅")
     else:
-        gr.Markdown("模型已加载，可以上传 CSV 或粘贴逗号分隔的一行特征进行预测。")
+        if MODEL_LOAD_ERROR:
+            gr.Markdown("**模型加载失败**，请查看运行日志（下方可能有堆栈）。")
+        else:
+            gr.Markdown("模型正在后台加载（不会阻塞应用启动），请稍候。")
     with gr.Row():
         file_in = gr.File(label="上传 CSV（每行 = 一个样本）")
         text_in = gr.Textbox(lines=2, placeholder="粘贴逗号分隔的一行特征，例如: 0.1,0.2,0.3,...")
@@ -128,36 +128,15 @@ with gr.Blocks() as demo:
 
     btn.click(run_predict, inputs=[file_in, text_in, n_ts, n_feat], outputs=[out_text, out_json])
 
-# Robust port selection & launch helper
-def find_free_port():
-    s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
-    s.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
-    s.bind(('', 0))
-    addr, port = s.getsockname()
-    s.close()
-    return port
-
-def get_desired_port():
-    # priority: GRADIO_SERVER_PORT -> PORT -> auto find
-    p = os.environ.get("GRADIO_SERVER_PORT") or os.environ.get("PORT")
-    if p:
-        try:
-            return int(p)
-        except:
-            pass
-    # fallback to ephemeral free port
-    return find_free_port()
-
-if __name__ == '__main__':
-    port = None
+# ---------------- Launch (Spaces-friendly) ----------------
+def get_port():
     try:
-        port = get_desired_port()
-        print(f"Launching server on port {port} (server_name=0.0.0.0)")
-        demo.launch(server_name='0.0.0.0', server_port=port)
-    except OSError as e:
-        print("Failed to bind requested port:", e)
-        print("Falling back to default demo.launch() (no explicit port).")
-        # last fallback: let Gradio choose/handle
-        # Colab / local debug: 生成外网临时链接（share=True）
-        demo.launch(share=True)
-
+        return int(os.environ.get("PORT", 7860))
+    except:
+        return 7860
+
+if __name__ == "__main__":
+    port = get_port()
+    print(f"[app] Starting Gradio on 0.0.0.0:{port}", flush=True)
+    # Do NOT use share=True on Spaces
+    demo.launch(server_name="0.0.0.0", server_port=port, show_error=True, enable_queue=True)