Update app.py

app.py

@@ -1,52 +1,33 @@
 import gradio as gr
 from openai import OpenAI
-from datasets import load_dataset, Dataset
+from datasets import Dataset
 from datetime import datetime, timedelta
 import pandas as pd
-import time, os, random, uuid, tempfile
+import time, os, random, uuid, tempfile, json
 
 # --- API設定 ---
-API_KEY = os.getenv("API_KEY")  # Hugging Face Secrets
+API_KEY = os.getenv("API_KEY")
 BASE_URL = "https://openrouter.ai/api/v1"
 HF_TOKEN = os.getenv("HF_TOKEN")
 DATASET_REPO = "Toya0421/lexile_test_logging"
 
-# --- ローカルログファイル ---
+# --- ログファイル ---
 LOG_FILE = "logs.csv"
 
-# --- Lexile難易度別教材 ---
-texts = {
-    300: [
-        "Tom has a red ball. He plays with it in the park. The sun is bright.",
-        "Anna likes cats. She feeds them milk every morning.",
-        "The boy sees a bird in the tree. It sings a happy song."
-    ],
-    600: [
-        "A young boy found a lost puppy near the river. He decided to take care of it.",
-        "Lisa visits her grandmother every weekend. They bake cookies together.",
-        "The farmer plants rice in the spring and harvests it in the autumn."
-    ],
-    850: [
-        "Sarah enjoyed reading stories about ancient civilizations and their discoveries.",
-        "Tom built a small robot that could move and pick up light objects.",
-        "Maria was fascinated by how airplanes could stay in the air for hours."
-    ],
-    1050: [
-        "The scientist developed a new hypothesis about the evolution of animal behavior.",
-        "A group of engineers designed a bridge that could resist strong winds.",
-        "Students discussed how technology changes communication in modern society."
-    ],
-    1250: [
-        "Philosophers have long debated the intricate relationship between free will and determinism.",
-        "Economists analyzed the impact of automation on global labor markets.",
-        "Researchers investigated how cultural identity influences moral decision-making."
-    ]
-}
+# --- passage本文は外部ファイルから読み込み ---
+PASSAGE_FILE = "passage.csv"
+passages_df = pd.read_csv(PASSAGE_FILE)
 
 levels = [300, 600, 850, 1050, 1250]
 
 client = OpenAI(base_url=BASE_URL, api_key=API_KEY)
 
+# --- passage取得 ---
+def get_passage(level):
+    candidates = passages_df[passages_df["lexile_level"] == level]
+    row = candidates.sample(1).iloc[0]
+    return row["passage_id"], row["text"]
+
 # --- 問題生成 ---
 def generate_question(text):
     prompt = f"""
@@ -73,17 +54,15 @@ def generate_question(text):
 def check_answer_with_ai(text, question, user_answer):
     prompt = f"""
     Read the passage and question below. Decide if the user's answer is correct.
-    Passage:
-    {text}
-    Question:
-    {question}
-    User Answer: {user_answer}
-    Respond with only one word: "Correct" or "Incorrect".
+    P: {text}
+    Q: {question}
+    User: {user_answer}
+    Respond only: Correct or Incorrect
     """
     response = client.chat.completions.create(
         model="google/gemma-3-27b-it:free",
         messages=[{"role": "user", "content": prompt}],
-        max_tokens=10,
+        max_tokens=15,
         temperature=0,
     )
     return response.choices[0].message.content.strip().lower() == "correct"
@@ -97,128 +76,130 @@ def adaptive_test(prev_level, prev_correct):
         return levels[idx - 1]
     return prev_level
 
-# --- ログ追記関数 ---
-def log_to_csv_and_push(user_id, level, passage, question, user_answer, correct, response_time):
-    entry = {
-        "user_id": user_id,
-        "lexile_level": level,
-        "passage": passage,
-        "question": question,
-        "user_answer": user_answer,
-        "correct": correct,
-        "response_time": response_time,
-        "timestamp": (datetime.utcnow() + timedelta(hours=9)).strftime("%Y-%m-%d %H:%M:%S"),
-    }
-
-    # --- ローカルに追記 ---
+# --- ログ書き込み（回答送信時のみ） ---
+def log_to_csv_and_push(entry):
     df = pd.DataFrame([entry])
     if os.path.exists(LOG_FILE):
         df.to_csv(LOG_FILE, mode="a", index=False, header=False)
     else:
         df.to_csv(LOG_FILE, index=False)
 
-    # --- データセットへ反映 ---
     all_logs = pd.read_csv(LOG_FILE)
-
     tmp_dir = tempfile.mkdtemp()
     tmp_path = os.path.join(tmp_dir, "data.parquet")
     all_logs.to_parquet(tmp_path)
-
     dataset = Dataset.from_parquet(tmp_path)
     dataset.push_to_hub(DATASET_REPO, token=HF_TOKEN)
 
-# --- 状態変数 ---
-used_texts = set()
+# --- 状態 ---
+used_passages = set()
 question_count = 0
 MAX_QUESTIONS = 5
 current_user_id = None
+action_log = []
 
-def start_test():
-    global used_texts, question_count, current_user_id
-    used_texts = set()
+# --- テスト開始 ---
+def start_test(user_input_id):
+    global used_passages, question_count, current_user_id, action_log
+    used_passages = set()
     question_count = 0
-    current_user_id = str(uuid.uuid4())
+    action_log = []
+
+    # ユーザーIDが空 → uuid割当
+    current_user_id = user_input_id.strip() if user_input_id else str(uuid.uuid4())
 
     level = 850
-    text = random.choice(texts[level])
-    used_texts.add(text)
+    passage_id, text = get_passage(level)
+    used_passages.add(passage_id)
+
     question = generate_question(text)
-    start_time = time.time()
-    return text, question, level, None, "", "", True, start_time, current_user_id
 
-def next_step(prev_level, user_answer, question_text, passage_text, start_time, user_id):
-    global question_count, used_texts
+    displayed_time = datetime.utcnow() + timedelta(hours=9)
+    action_log.append({"action": "displayed", "time": displayed_time.isoformat(), "question_number": 1})
 
-    if user_answer is None or user_answer == "":
-        return "⚠️ Please select an answer!", passage_text, question_text, prev_level, None, "", True, start_time, user_id
+    return text, question, level, passage_id, "", "", True, displayed_time.isoformat(), current_user_id
 
-    question_count += 1
+# --- 回答送信 ---
+def next_step(prev_level, user_answer, question_text, passage_text, displayed_time, user_id, passage_id):
+    global question_count, used_passages, action_log
+
+    if not user_answer:
+        return "⚠️ Please select an answer!", passage_text, question_text, prev_level, None, "", True, displayed_time, user_id, passage_id
 
-    end_time = time.time()
-    response_time = round(end_time - start_time, 2)
+    question_count += 1
+    submit_time = datetime.utcnow() + timedelta(hours=9)
     correct = check_answer_with_ai(passage_text, question_text, user_answer)
-    new_level = adaptive_test(prev_level, correct)
 
-    # ログ追記
-    log_to_csv_and_push(user_id, prev_level, passage_text, question_text, user_answer, correct, response_time)
+    # ログ書き込み（回答送信時のみ）
+    entry = {
+        "user_id": user_id,
+        "question_number": question_count,
+        "lexile_level": prev_level,
+        "passage_id": passage_id,
+        "question": question_text,
+        "user_answer": user_answer,
+        "correct": correct,
+        "displayed_time": displayed_time,
+        "submit_time": submit_time.strftime("%Y-%m-%d %H:%M:%S"),
+        "actions": json.dumps(action_log, ensure_ascii=False),
+    }
+    log_to_csv_and_push(entry)
 
+    new_level = adaptive_test(prev_level, correct)
     feedback = "✅ Correct!" if correct else "❌ Incorrect."
 
+    # 終了
     if question_count >= MAX_QUESTIONS:
-        feedback = f"🎯 **Test finished!**\n\nYour estimated reading level is **{new_level}L**.\n\nThank you for participating!"
-        return feedback, "", "", new_level, None, "", False, 0.0, user_id
+        return f"🎯 Test finished!\nYour reading level: **{new_level}L**", "", "", new_level, None, "", False, "", user_id, passage_id
 
-    available = [t for t in texts[new_level] if t not in used_texts]
+    # 次の問題へ
+    candidates = passages_df[passages_df["lexile_level"] == new_level]
+    available = [pid for pid in candidates["passage_id"] if pid not in used_passages]
     if not available:
-        available = texts[new_level]
-    next_text = random.choice(available)
-    used_texts.add(next_text)
+        available = list(candidates["passage_id"])
+
+    next_passage_id = random.choice(available)
+    next_text = passages_df[passages_df["passage_id"] == next_passage_id]["text"].iloc[0]
+    used_passages.add(next_passage_id)
+
     next_question = generate_question(next_text)
-    feedback += "\n➡️ Next question loading..."
-    next_start_time = time.time()
-    return feedback, next_text, next_question, new_level, None, "", True, next_start_time, user_id
+    next_display_time = datetime.utcnow() + timedelta(hours=9)
+
+    action_log = [{"action": "displayed", "time": next_display_time.isoformat(), "question_number": question_count+1}]
+
+    return feedback + "\n➡️ Loading next question…", next_text, next_question, new_level, None, "", True, next_display_time.isoformat(), user_id, next_passage_id
 
 # --- Gradio UI ---
 with gr.Blocks() as demo:
-    gr.Markdown("# 📘 Lexile Level Test")
+    gr.Markdown("# 📘 Lexile Adaptive Reading Test")
 
+    user_input_id = gr.Textbox(label="Enter your Student ID", placeholder="e.g. 2025012")
     start_btn = gr.Button("▶️ Start Test")
 
-    text_display = gr.Textbox(label="Reading Passage", lines=6, interactive=False, visible=True)
-    question_display = gr.Textbox(label="Generated Question", lines=8, interactive=False, visible=True)
-    user_answer = gr.Radio(choices=["A", "B", "C", "D"], label="Your Answer", visible=True)
-    submit_btn = gr.Button("Submit Answer", visible=True)
+    text_display = gr.Textbox(label="Reading Passage", lines=6, interactive=False)
+    question_display = gr.Textbox(label="Question", lines=7, interactive=False)
+    answer_box = gr.Radio(choices=["A", "B", "C", "D"], label="Select Answer")
+    submit_btn = gr.Button("Submit Answer")
 
     feedback_display = gr.Markdown()
     hidden_level = gr.Number(visible=False)
-    hidden_passage = gr.Textbox(visible=False)
+    hidden_passage_id = gr.Textbox(visible=False)
     test_visible = gr.State(True)
-    hidden_start_time = gr.Number(visible=False)
+    hidden_display_time = gr.Textbox(visible=False)
     hidden_user_id = gr.Textbox(visible=False)
 
     start_btn.click(
         fn=start_test,
-        inputs=[],
-        outputs=[text_display, question_display, hidden_level, user_answer, feedback_display,
-                 hidden_passage, test_visible, hidden_start_time, hidden_user_id]
+        inputs=[user_input_id],
+        outputs=[text_display, question_display, hidden_level, hidden_passage_id, answer_box,
+                 feedback_display, test_visible, hidden_display_time, hidden_user_id]
     )
 
     submit_btn.click(
         fn=next_step,
-        inputs=[hidden_level, user_answer, question_display, text_display, hidden_start_time, hidden_user_id],
-        outputs=[feedback_display, text_display, question_display, hidden_level, user_answer,
-                 hidden_passage, test_visible, hidden_start_time, hidden_user_id]
-    )
-
-    def toggle_visibility(show):
-        v = bool(show)
-        return (gr.update(visible=v), gr.update(visible=v),
-                gr.update(visible=v), gr.update(visible=v))
-
-    test_visible.change(
-        fn=toggle_visibility,
-        inputs=test_visible,
-        outputs=[text_display, question_display, user_answer, submit_btn]
+        inputs=[hidden_level, answer_box, question_display, text_display, hidden_display_time, hidden_user_id, hidden_passage_id],
+        outputs=[feedback_display, text_display, question_display, hidden_level, answer_box,
+                 hidden_passage_id, test_visible, hidden_display_time, hidden_user_id, hidden_passage_id]
     )
 
 demo.launch()