Update app.py

app.py

@@ -3,9 +3,11 @@ from openai import OpenAI
 from datasets import Dataset
 from datetime import datetime, timedelta
 import pandas as pd
-import time, os, random, tempfile, json, glob
+import time, os, random, tempfile, json, glob, re
 
-# --- API / HF 設定 ---
+# ======================================================
+# API / HF 設定
+# ======================================================
 API_KEY = os.getenv("API_KEY")
 BASE_URL = "https://openrouter.ai/api/v1"
 HF_TOKEN = os.getenv("HF_TOKEN")
@@ -14,23 +16,25 @@ LOG_FILE = "reading_logs.csv"
 
 client = OpenAI(base_url=BASE_URL, api_key=API_KEY)
 
-# --- passage_information.xlsx 読み込み (Text# と flesch_score 使用) ---
+# ======================================================
+# passage_information.xlsx 読み込み
+# ======================================================
 passage_info_df = pd.read_excel("passage_information.xlsx")
 
-# --- 状態変数 ---
+# ======================================================
+# グローバル状態（※最小限）
+# ======================================================
 used_passages = set()
 current_user_id = None
 current_level = None
 
+# rewrite キャッシュ
+rewrite_cache = {}
 
 # ======================================================
-#  新しい教材管理：passages フォルダからランダム選択
+# 教材ロード
 # ======================================================
-
 def load_passage_file(text_id):
-    """
-    passages/pg{text_id}.txt を読み込み、内容を返す。
-    """
     path = f"passages/pg{text_id}.txt"
     if not os.path.exists(path):
         return None
@@ -38,124 +42,97 @@ def load_passage_file(text_id):
         return f.read()
 
 def get_new_passage_random():
-    """
-    passages フォルダからランダムに教材を選び（pg◯.txt）、
-    passage_information.xlsx の Text# の flesch_score を original_level として返す。
-    """
     global used_passages
-
-    # --- pg*.txt を取得 ---
     files = glob.glob("passages/pg*.txt")
-    if not files:
-        return None, None, None
-
-    # --- ファイル名から Text# (整数) を抽出 ---
-    all_ids = []
-    for f in files:
-        name = os.path.basename(f)
-        num = name.replace("pg", "").replace(".txt", "")
-        if num.isdigit():
-            all_ids.append(int(num))
-
-    # --- 未使用の ID を優先 ---
+    all_ids = [int(os.path.basename(f).replace("pg", "").replace(".txt", ""))
+               for f in files]
+
     available = [pid for pid in all_ids if pid not in used_passages]
     if not available:
         used_passages.clear()
-        available = list(all_ids)
+        available = all_ids
 
-    # --- ランダムに選択 ---
-    text_id = random.choice(available)
-    used_passages.add(text_id)
+    pid = random.choice(available)
+    used_passages.add(pid)
 
-    # --- テキスト読み込み ---
-    text = load_passage_file(text_id)
-    if text is None:
-        return None, None, None
+    text = load_passage_file(pid)
+    row = passage_info_df[passage_info_df["Text#"] == pid]
+    orig_level = row.iloc[0]["flesch_score"] if len(row) else None
 
-    # --- Excel から original_level (flesch_score) を取得 ---
-    row = passage_info_df[passage_info_df["Text#"] == text_id]
-    if len(row) == 0:
-        orig_level = None
-    else:
-        orig_level = row.iloc[0]["flesch_score"]
+    return pid, text, orig_level
 
-    return text_id, text, orig_level
+# ======================================================
+# ページ分割
+# ======================================================
+def split_pages(text, max_words=300):
+    sentences = re.split(r'(?<=[.!?])\s+', text.strip())
+    pages, buf, wc = [], [], 0
+    for s in sentences:
+        w = len(s.split())
+        if wc + w > max_words:
+            pages.append(" ".join(buf))
+            buf, wc = [s], w
+        else:
+            buf.append(s)
+            wc += w
+    if buf:
+        pages.append(" ".join(buf))
+    return pages
 
+# ======================================================
+# 文脈要約（軽量）
+# ======================================================
+def summarize_for_context(text, max_sentences=2):
+    sents = re.split(r'(?<=[.!?])\s+', text.strip())
+    return " ".join(sents[:max_sentences])
 
 # ======================================================
-#  Rewrite
+# 文脈付きページ rewrite（キャッシュあり）
 # ======================================================
+def rewrite_page_with_context(pid, page_index, original_pages, target_level):
+    key = f"{pid}_{target_level}_{page_index}"
+    if key in rewrite_cache:
+        return rewrite_cache[key]
 
-def rewrite_level(text, target_level):
-    level_to_flesch = {
-        1: 90,
-        2: 70,
-        3: 55,
-        4: 40,
-        5: 25
-    }
+    prev_summary = summarize_for_context(original_pages[page_index - 1]) if page_index > 0 else ""
+    next_summary = summarize_for_context(original_pages[page_index + 1]) if page_index < len(original_pages) - 1 else ""
 
+    level_to_flesch = {1:90, 2:70, 3:55, 4:40, 5:25}
     target_flesch = level_to_flesch[int(target_level)]
 
     prompt = f"""
-Rewrite the following passage so it fits about {target_flesch} Flesch Reading Ease Score
-- Extract only the portions of the text that should be read as the main body,
-excluding the title, author name, source information, chapter number, annotations, and footers.
-- When outputting, make sure sections divided by chapters, etc., are clearly distinguishable by leaving a blank line between them.
-- Preserve the original meaning faithfully. 
-- Do not add new information or remove essential information. 
-- Output only the rewritten passage. Do not include explanations. 
-{text}
+You are rewriting ONE page of a passage.
+
+Context (for coherence only):
+- Previous summary:
+{prev_summary}
+
+- Next summary:
+{next_summary}
+
+Rewrite ONLY the following page so that it fits about {target_flesch} Flesch Reading Ease.
+- Preserve meaning and references.
+- Do not add or remove information.
+- Output only the rewritten page.
+
+[Page]
+{original_pages[page_index]}
 """
 
     resp = client.chat.completions.create(
         model="google/gemini-2.5-flash",
         messages=[{"role": "user", "content": prompt}],
         temperature=0.4,
-        max_tokens=5000
+        max_tokens=1200
     )
-    return resp.choices[0].message.content.strip()
-
-
-import re
-
-def split_pages(text, max_words=300):
-    """
-    文単位でページを分割する。
-    - 文の途中でページを分割しない
-    - max_words の上限を超えないようにする
-    """
-    # 文に分割（. ? ! のあとに改行やスペースが続くパターン）
-    sentences = re.split(r'(?<=[.!?])\s+', text.strip())
-    pages = []
-    current_page = []
-    current_word_count = 0
-
-    for sentence in sentences:
-        words = sentence.split()
-        sentence_len = len(words)
-
-        # 次の文を追加すると max_words を超える場合 → 新しいページを作る
-        if current_word_count + sentence_len > max_words:
-            if current_page:
-                pages.append(" ".join(current_page))
-            current_page = [sentence]
-            current_word_count = sentence_len
-        else:
-            current_page.append(sentence)
-            current_word_count += sentence_len
-
-    # 最後のページを追加
-    if current_page:
-        pages.append(" ".join(current_page))
-
-    return pages or [text]
 
+    rewritten = resp.choices[0].message.content.strip()
+    rewrite_cache[key] = rewritten
+    return rewritten
 
 # ======================================================
-#  Save Log
+# ログ保存
 # ======================================================
-
 def save_log(entry):
     df = pd.DataFrame([entry])
     if os.path.exists(LOG_FILE):
@@ -163,459 +140,133 @@ def save_log(entry):
     else:
         df.to_csv(LOG_FILE, index=False)
 
-    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)
-
-
 # ======================================================
-#  Start
+# Start
 # ======================================================
-
 def start_test(student_id, level_input):
     global current_user_id, current_level, used_passages
     used_passages = set()
 
-    action = "start_pushed"
-    now = (datetime.utcnow() + timedelta(hours=9)).isoformat()
+    if not student_id:
+        return "", "", "", "", 0, "", "", None, None, None
 
-    entry = {
-        "user_id": student_id,
-        "assigned_level": current_level,
-        "passage_id": None,
-        "original_level": None,
-        "action_time": now,
-        "action_type": action,
-        "page_text": None
-    }
-    save_log(entry)
-
-    if not student_id or str(student_id).strip() == "":
-        return (
-            "", "", json.dumps([]), 0, "",
-            0, "", None, None,
-            gr.update(interactive=False, visible=False),
-            gr.update(interactive=False, visible=True),
-            gr.update(interactive=False, visible=False)
-        )
-
-    current_user_id = str(student_id).strip()
+    current_user_id = student_id
     current_level = int(level_input)
 
     pid, text, orig_lev = get_new_passage_random()
-    if text is None:
-        return (
-            "教材が見つかりません", "", json.dumps([]), 0, "",
-            0, "", None, None,
-            gr.update(interactive=False, visible=False),
-            gr.update(interactive=False, visible=False),
-            gr.update(interactive=False, visible=False)
-        )
-
-    rewritten = rewrite_level(text, current_level)
-    pages = split_pages(rewritten)
-    total = len(pages)
-
-    if total == 1:
-        prev_upd = gr.update(interactive=False, visible=False)
-        next_upd = gr.update(interactive=False, visible=False)
-        finish_upd = gr.update(interactive=True, visible=True)
-    else:
-        prev_upd = gr.update(interactive=False, visible=False)
-        next_upd = gr.update(interactive=True, visible=True)
-        finish_upd = gr.update(interactive=False, visible=False)
+    original_pages = split_pages(text)
 
-    page_num = 1
-    now = (datetime.utcnow() + timedelta(hours=9)).isoformat()
+    rewritten_pages = [None] * len(original_pages)
+    rewritten_pages[0] = rewrite_page_with_context(
+        pid, 0, original_pages, current_level
+    )
 
-    entry = {
-        "user_id": current_user_id,
+    save_log({
+        "user_id": student_id,
         "assigned_level": current_level,
         "passage_id": pid,
         "original_level": orig_lev,
-        "action_time": now,
-        "action_type": f"page_displayed_{page_num}",
-        "page_text": pages[0]
-    }
-    save_log(entry)
+        "action_time": datetime.now().isoformat(),
+        "action_type": "page_displayed_1",
+        "page_text": rewritten_pages[0]
+    })
 
     return (
-        pages[0],
-        f"1 / {total}",
-        json.dumps(pages, ensure_ascii=False),
+        rewritten_pages[0],
+        f"1 / {len(original_pages)}",
+        json.dumps(rewritten_pages, ensure_ascii=False),
+        json.dumps(original_pages, ensure_ascii=False),
         0,
-        total,
         pid,
         orig_lev,
-        current_level,
-        prev_upd,
-        next_upd,
-        finish_upd
+        gr.update(interactive=False, visible=False),
+        gr.update(interactive=len(original_pages)>1, visible=len(original_pages)>1),
+        gr.update(interactive=len(original_pages)==1, visible=len(original_pages)==1)
     )
 
-
 # ======================================================
-#  Next / Prev / Finish（以下は元コードのまま）
+# Next / Prev
 # ======================================================
-
-def next_page(pages_json, current_page, total_pages, pid, orig_lev):
-    now = (datetime.utcnow() + timedelta(hours=9)).isoformat()
-
-    entry = {
-        "user_id": current_user_id,
-        "assigned_level": current_level,
-        "passage_id": pid,
-        "original_level": orig_lev,
-        "action_time": now,
-        "action_type": "next_pushed",
-        "page_text": None
-    }
-    save_log(entry)
-
+def next_page(pages_json, original_json, page_idx, pid, orig_lev):
     pages = json.loads(pages_json)
-    if not pages:
-        return ("", "", json.dumps([]), 0,
-                gr.update(interactive=False, visible=False),
-                gr.update(interactive=False, visible=False),
-                gr.update(interactive=False, visible=False))
+    original = json.loads(original_json)
+    idx = page_idx + 1
 
-    new_page = min(current_page + 1, total_pages - 1)
+    if pages[idx] is None:
+        pages[idx] = rewrite_page_with_context(pid, idx, original, current_level)
 
-    now2 = (datetime.utcnow() + timedelta(hours=9)).isoformat()
-    entry2 = {
+    save_log({
         "user_id": current_user_id,
         "assigned_level": current_level,
         "passage_id": pid,
         "original_level": orig_lev,
-        "action_time": now2,
-        "action_type": f"page_displayed_{new_page+1}",
-        "page_text": pages[new_page]
-    }
-    save_log(entry2)
-
-    if new_page == total_pages - 1:
-        return (
-            pages[new_page],
-            f"{new_page+1} / {total_pages}",
-            json.dumps(pages),
-            new_page,
-            gr.update(interactive=True, visible=True),
-            gr.update(interactive=False, visible=False),
-            gr.update(interactive=True, visible=True)
-        )
+        "action_time": datetime.now().isoformat(),
+        "action_type": f"page_displayed_{idx+1}",
+        "page_text": pages[idx]
+    })
 
+    total = len(original)
     return (
-        pages[new_page],
-        f"{new_page+1} / {total_pages}",
-        json.dumps(pages),
-        new_page,
-        gr.update(interactive=(new_page > 0), visible=(new_page > 0)),
-        gr.update(interactive=True, visible=True),
-        gr.update(interactive=False, visible=False)
+        pages[idx],
+        f"{idx+1} / {total}",
+        json.dumps(pages, ensure_ascii=False),
+        idx,
+        gr.update(interactive=idx>0, visible=idx>0),
+        gr.update(interactive=idx<total-1, visible=idx<total-1),
+        gr.update(interactive=idx==total-1, visible=idx==total-1)
     )
 
-
-def prev_page(pages_json, current_page, total_pages, pid, orig_lev):
-    now = (datetime.utcnow() + timedelta(hours=9)).isoformat()
-
-    entry = {
-        "user_id": current_user_id,
-        "assigned_level": current_level,
-        "passage_id": pid,
-        "original_level": orig_lev,
-        "action_time": now,
-        "action_type": "prev_pushed",
-        "page_text": None
-    }
-    save_log(entry)
-
+def prev_page(pages_json, page_idx):
     pages = json.loads(pages_json)
-    if not pages:
-        return ("", "", json.dumps([]), 0,
-                gr.update(interactive=False, visible=False),
-                gr.update(interactive=False, visible=False),
-                gr.update(interactive=False, visible=False))
-
-    new_page = max(current_page - 1, 0)
-
-    prev_upd = gr.update(interactive=(new_page > 0), visible=(new_page > 0))
-    next_visible = (new_page < total_pages - 1)
-    next_upd = gr.update(interactive=next_visible, visible=next_visible)
-    finish_upd = gr.update(interactive=(not next_visible), visible=(not next_visible))
-
-    now2 = (datetime.utcnow() + timedelta(hours=9)).isoformat()
-    entry2 = {
-        "user_id": current_user_id,
-        "assigned_level": current_level,
-        "passage_id": pid,
-        "original_level": orig_lev,
-        "action_time": now2,
-        "action_type": f"page_displayed_{new_page+1}",
-        "page_text": pages[new_page]
-    }
-    save_log(entry2)
-
+    idx = page_idx - 1
     return (
-        pages[new_page],
-        f"{new_page+1} / {total_pages}",
-        json.dumps(pages),
-        new_page,
-        prev_upd,
-        next_upd,
-        finish_upd
-    )
-
-
-def finish_or_retire(pages_json, current_page, pid, orig_lev, action):
-    pages = json.loads(pages_json)
-    now = (datetime.utcnow() + timedelta(hours=9)).isoformat()
-
-    entry = {
-        "user_id": current_user_id,
-        "assigned_level": current_level,
-        "passage_id": pid,
-        "original_level": orig_lev,
-        "action_time": now,
-        "action_type": action,
-        "page_text": None
-    }
-    save_log(entry)
-
-    new_pid, new_text, new_orig_lev = get_new_passage_random()
-    if new_text is None:
-        return (
-            "教材がありません", "", json.dumps([]), 0, "",
-            0, "", None, None,
-            gr.update(interactive=False, visible=False),
-            gr.update(interactive=False, visible=False),
-            gr.update(interactive=False, visible=False)
-        )
-
-    rewritten = rewrite_level(new_text, current_level)
-    new_pages = split_pages(rewritten)
-    total = len(new_pages)
-
-    if total == 1:
-        prev_upd = gr.update(interactive=False, visible=False)
-        next_upd = gr.update(interactive=False, visible=False)
-        finish_upd = gr.update(interactive=True, visible=True)
-    else:
-        prev_upd = gr.update(interactive=False, visible=False)
-        next_upd = gr.update(interactive=True, visible=True)
-        finish_upd = gr.update(interactive=False, visible=False)
-
-    now2 = (datetime.utcnow() + timedelta(hours=9)).isoformat()
-    entry2 = {
-        "user_id": current_user_id,
-        "assigned_level": current_level,
-        "passage_id": new_pid,
-        "original_level": new_orig_lev,
-        "action_time": now2,
-        "action_type": "page_displayed_1",
-        "page_text": new_pages[0]
-    }
-    save_log(entry2)
-
-    return (
-        new_pages[0],
-        f"1 / {total}",
-        json.dumps(new_pages, ensure_ascii=False),
-        0,
-        total,
-        new_pid,
-        new_orig_lev,
-        current_level,
-        prev_upd,
-        next_upd,
-        finish_upd
+        pages[idx],
+        "",
+        pages_json,
+        idx,
+        gr.update(interactive=idx>0, visible=idx>0),
+        gr.update(interactive=True, visible=True),
+        gr.update(interactive=False, visible=False)
     )
 
-
 # ======================================================
-#  UI
+# UI
 # ======================================================
-custom_css = """
-/* ===============================
-   共通（両モード）
-   =============================== */
-.big-text {
-    font-size: 22px !important;
-    line-height: 1.8 !important;
-    font-family: "Noto Sans", sans-serif !important;
-}
-
-/* 教材表示ボックス */
-.reading-area {
-    padding: 20px !important;
-    border-radius: 12px !important;
-    border: 1px solid #ccc !important;
-    transition: background-color 0.2s ease, color 0.2s ease;
-}
-
-/* ===============================
-   ライトモード
-   =============================== */
-@media (prefers-color-scheme: light) {
-    body, .gradio-container {
-        background-color: #ffffff !important;
-        color: #222 !important;
-    }
-
-    .reading-area {
-        background-color: #fafafa !important;
-        color: #222 !important;
-        border-color: #ddd !important;
-    }
-
-    textarea, input, .gr-textbox textarea {
-        background-color: #ffffff !important;
-        color: #222 !important;
-        border: 1px solid #ccc !important;
-    }
-}
-
-/* ===============================
-   ダークモード
-   =============================== */
-@media (prefers-color-scheme: dark) {
-    body, .gradio-container {
-        background-color: #1e1e1e !important;
-        color: #e6e6e6 !important;
-    }
-
-    /* 教材の背景は黒すぎると読みにくいのでやや明るめのチャコール */
-    .reading-area {
-        background-color: #2a2a2a !important;
-        color: #f2f2f2 !important;
-        border-color: #444 !important;
-    }
-
-    textarea, input, .gr-textbox textarea {
-        background-color: #2c2c2c !important;
-        color: #f0f0f0 !important;
-        border: 1px solid #555 !important;
-    }
-
-    /* ボタン��見やすく */
-    button {
-        background-color: #3a3a3a !important;
-        color: #f0f0f0 !important;
-        border: 1px solid #555 !important;
-    }
-    button:hover {
-        background-color: #4a4a4a !important;
-    }
-
-    .gr-panel, .gr-box, .gr-group {
-        background-color: #272727 !important;
-        border-color: #444 !important;
-    }
-}
-"""
-
-
-with gr.Blocks(css=custom_css) as demo:
+with gr.Blocks() as demo:
     gr.Markdown("# 📚 Reading Exercise")
 
-    student_id_input = gr.Textbox(label="学生番号（必須）")
-    level_input = gr.Dropdown(
-        choices=[1,2,3,4,5],
-        label="あなたの Reading Level（Level Testの結果を選択）",
-        value=3
-    )
-
-    start_btn = gr.Button("スタート")
+    student_id = gr.Textbox(label="学生番号")
+    level = gr.Dropdown([1,2,3,4,5], value=3, label="Reading Level")
+    start = gr.Button("スタート")
 
-    text_display = gr.Textbox(
-        label="教材",
-        lines=18,
-        interactive=False,
-        elem_classes=["big-text", "reading-area"]
-    )
-    page_display = gr.Textbox(label="進行状況", lines=1, interactive=False)
+    text = gr.Textbox(lines=18, interactive=False)
+    page = gr.Textbox(interactive=False)
 
     hidden_pages = gr.Textbox(visible=False)
-    hidden_page_index = gr.Number(visible=False)
-    hidden_total_pages = gr.Number(visible=False)
-    hidden_passage_id = gr.Textbox(visible=False)
+    hidden_original = gr.Textbox(visible=False)
+    hidden_idx = gr.Number(visible=False)
+    hidden_pid = gr.Textbox(visible=False)
     hidden_orig_lev = gr.Textbox(visible=False)
-    hidden_assigned_lev = gr.Textbox(visible=False)
 
     with gr.Row():
-        prev_btn = gr.Button("◀ 前へ", interactive=False, visible=False)
-        next_btn = gr.Button("次へ ▶", interactive=False, visible=False)
-        finish_btn = gr.Button("読み終えた", interactive=False, visible=False)
-
-    retire_btn = gr.Button("リタイア")
-
-    start_btn.click(
-        fn=start_test,
-        inputs=[student_id_input, level_input],
-        outputs=[
-            text_display, page_display,
-            hidden_pages, hidden_page_index,
-            hidden_total_pages, hidden_passage_id,
-            hidden_orig_lev, hidden_assigned_lev,
-            prev_btn, next_btn, finish_btn
-        ]
+        prev_btn = gr.Button("◀ 前へ", visible=False)
+        next_btn = gr.Button("次へ ▶", visible=False)
+        finish_btn = gr.Button("読み終えた", visible=False)
+
+    start.click(
+        start_test,
+        [student_id, level],
+        [text, page, hidden_pages, hidden_original,
+         hidden_idx, hidden_pid, hidden_orig_lev,
+         prev_btn, next_btn, finish_btn]
     )
 
     next_btn.click(
-        fn=next_page,
-        inputs=[
-            hidden_pages, hidden_page_index,
-            hidden_total_pages, hidden_passage_id,
-            hidden_orig_lev
-        ],
-        outputs=[
-            text_display, page_display,
-            hidden_pages, hidden_page_index,
-            prev_btn, next_btn, finish_btn
-        ]
-    )
-
-    prev_btn.click(
-        fn=prev_page,
-        inputs=[
-            hidden_pages, hidden_page_index,
-            hidden_total_pages, hidden_passage_id,
-            hidden_orig_lev
-        ],
-        outputs=[
-            text_display, page_display,
-            hidden_pages, hidden_page_index,
-            prev_btn, next_btn, finish_btn
-        ]
-    )
-
-    finish_btn.click(
-        fn=lambda p, i, pid, o: finish_or_retire(p, i, pid, o, "finished"),
-        inputs=[hidden_pages, hidden_page_index, hidden_passage_id, hidden_orig_lev],
-        outputs=[
-            text_display, page_display,
-            hidden_pages, hidden_page_index,
-            hidden_total_pages, hidden_passage_id,
-            hidden_orig_lev, hidden_assigned_lev,
-            prev_btn, next_btn, finish_btn
-        ]
-    )
-
-    retire_btn.click(
-        fn=lambda p, i, pid, o: finish_or_retire(p, i, pid, o, "retire"),
-        inputs=[
-            hidden_pages, hidden_page_index,
-            hidden_passage_id, hidden_orig_lev
-        ],
-        outputs=[
-            text_display, page_display,
-            hidden_pages, hidden_page_index,
-            hidden_total_pages, hidden_passage_id,
-            hidden_orig_lev, hidden_assigned_lev,
-            prev_btn, next_btn, finish_btn
-        ]
+        next_page,
+        [hidden_pages, hidden_original, hidden_idx,
+         hidden_pid, hidden_orig_lev],
+        [text, page, hidden_pages, hidden_idx,
+         prev_btn, next_btn, finish_btn]
     )
 
 demo.launch()