Update app.py

app.py

@@ -1,206 +1,319 @@
 import gradio as gr
-import pandas as pd
-import datetime
-import json
-import uuid
-import os
-from datasets import load_dataset, Dataset
 from openai import OpenAI
+from datasets import Dataset
+from datetime import datetime, timedelta
+import pandas as pd
+import time, os, random, uuid, tempfile, json
 
+# --- API設定 ---
 API_KEY = os.getenv("API_KEY")
+BASE_URL = "https://openrouter.ai/api/v1"
 HF_TOKEN = os.getenv("HF_TOKEN")
-HF_DATASET = "Toya0421/reading_exercise_logging"
+DATASET_REPO = "Toya0421/reading_exercise_logging"
+LOG_FILE = "reading_logs.csv"
 
-client = OpenAI(api_key=API_KEY)
+client = OpenAI(base_url=BASE_URL, api_key=API_KEY)
 
-# ✅ passage.csv 読み込み
-if os.path.exists("passage.csv") and os.path.getsize("passage.csv") > 0:
-    passages_df = pd.read_csv("passage.csv")
-else:
-    passages_df = pd.DataFrame(columns=["passage_id","genre","text","original_lexile_score"])
+# --- passage.csv 読み込み ---
+# (columns: passage_id, genre, text, original_lexile_score)
+passages_df = pd.read_csv("passage.csv")
+
+genres = [
+    "Literature","Science＆Technology","History","Social Science＆Society",
+    "Arts＆Culture","Religion＆Philosophy","Lifestyle＆Hobbies",
+    "Health＆Medicine","Education＆Reference"
+]
 
-# ✅ ページ分割（1ページ=約120ワード）
-def split_into_pages(text, words_per_page=120):
-    words = text.split()
-    pages = [" ".join(words[i:i+words_per_page]) for i in range(0, len(words), words_per_page)]
-    return pages
+# --- 状態 ---
+used_passages = set()
+current_user_id = None
+current_genre = None
+current_lexile = None
+action_log = []  # next/backなどページ操作ログ保存
 
-# ✅ 書き換え
+# --- 書き換え ---
 def rewrite_to_lexile(text, target_lexile):
     prompt = f"""
-You are an expert at leveling reading materials based on Lexile measures.
-
-Rewrite the following English passage so that the estimated Lexile score is close to {target_lexile}.
-- Keep original meaning.
-- Keep length similar.
-- Use vocabulary and structure appropriate for the target Lexile.
-
-Return ONLY the rewritten passage. Do not add explanations.
+Rewrite the following passage so it fits about {target_lexile} Lexile.
+- Keep original meaning and length
+- Avoid figurative language
+- Use simple syntax
+- Output only the rewritten passage
 
 {text}
 """
     resp = client.chat.completions.create(
-        model="gpt-5",
+        model="google/gemma-3-27b-it:free",
         messages=[{"role": "user", "content": prompt}],
-        temperature=0.3,
+        temperature=0.4,
+        max_tokens=1000
     )
     return resp.choices[0].message.content.strip()
 
-# ✅ Hugging Faceにログ保存
-def save_log_to_hf(row):
-    try:
-        dataset = load_dataset(HF_DATASET, split="train")
-        dataset = dataset.add_item(row)
-        dataset.push_to_hub(HF_DATASET)
-    except:
-        dataset = Dataset.from_list([row])
-        dataset.push_to_hub(HF_DATASET)
-
-# ✅ ローカルにも追記
-def save_local_csv(row):
-    df = pd.DataFrame([row])
-    if os.path.exists("full_log.csv"):
-        df.to_csv("full_log.csv", mode="a", header=False, index=False)
+# --- ページ分割 ---
+def split_pages(text, words=120):
+    w = text.split()
+    return [" ".join(w[i:i+words]) for i in range(0, len(w), words)]
+
+# --- next教材取得（同ジャンル or 別ジャンル） ---
+def get_new_passage(same_genre=True):
+    global used_passages, current_genre
+
+    if same_genre:
+        df = passages_df[passages_df["genre"] == current_genre]
+    else:
+        df = passages_df[passages_df["genre"] != current_genre]
+
+    if len(df) == 0:
+        return None, None, None
+
+    available = [pid for pid in df["passage_id"] if pid not in used_passages]
+
+    # 使い切ったらリセット
+    if not available:
+        used_passages.clear()
+        available = list(df["passage_id"])
+
+    pid = random.choice(available)
+    row = df[df["passage_id"] == pid].iloc[0]
+    used_passages.add(pid)
+
+    return pid, row["text"], row["original_lexile_score"]
+
+# --- ログ保存 ---
+def save_log(entry):
+    df = pd.DataFrame([entry])
+
+    # ✅ CSV追記
+    if os.path.exists(LOG_FILE):
+        df.to_csv(LOG_FILE, mode="a", index=False, header=False)
     else:
-        df.to_csv("full_log.csv", index=False)
-
-# ✅ 状態管理
-session_state = {
-    "user_id": None,
-    "genre": None,
-    "lexile": None,
-    "passage_id": None,
-    "original_lexile": None,
-    "pages": [],
-    "current_page": 0,
-    "actions": [],
-    "start_time": None,
-}
-
-# ✅ テスト開始
+        df.to_csv(LOG_FILE, index=False)
+
+    # ✅ HuggingFaceへpush
+    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)
+
+# =========================
+#  Gradio動作
+# =========================
+
+# --- スタート ---
 def start_test(student_id, genre, lexile):
-    if not student_id:
-        return "学生番号を入力してください", "", "", gr.update(interactive=False)
+    global current_user_id, current_genre, current_lexile, used_passages, action_log
+    used_passages = set()
+    action_log = []
 
-    session_state["user_id"] = student_id
-    session_state["genre"] = genre
-    session_state["lexile"] = int(lexile)
+    if not student_id or student_id.strip() == "":
+        return (
+            "", "", "", 0,
+            "⚠️ 学生番号を入力してください", False, "", 0, "", "", 0
+        )
 
-    row = passages_df[passages_df["genre"] == genre].sample(1).iloc[0]
-    pid = row["passage_id"]
-    orig_text = row["text"]
-    orig_lex = row["original_lexile_score"]
+    current_user_id = student_id.strip()
+    current_genre = genre
+    current_lexile = int(lexile)
 
-    rewritten = rewrite_to_lexile(orig_text, session_state["lexile"])
-    pages = split_into_pages(rewritten)
+    pid, text, orig_lex = get_new_passage(same_genre=True)
+    if text is None:
+        return ("", "", "", 0, "教材がありません", False, "", 0, "", "", 0)
 
-    session_state.update({
-        "passage_id": pid,
-        "original_lexile": orig_lex,
-        "pages": pages,
-        "current_page": 0,
-        "start_time": datetime.datetime.now().isoformat(),
-        "actions": ["first_page_displayed"]
-    })
+    rewritten = rewrite_to_lexile(text, current_lexile)
+    pages = split_pages(rewritten)
+
+    start_time = (datetime.utcnow() + timedelta(hours=9)).isoformat()
 
     return (
         pages[0],
-        f"1 / {len(pages)} ページ",
-        "",
-        gr.update(interactive=True)
+        f"1 / {len(pages)}",
+        json.dumps(pages, ensure_ascii=False),
+        0,  # current_page
+        "",  # feedback
+        True,  # 表示ON
+        start_time,
+        len(pages),
+        pid,
+        orig_lex,
+        current_lexile
     )
 
-def next_page():
-    if session_state["current_page"] < len(session_state["pages"]) - 1:
-        session_state["current_page"] += 1
-        session_state["actions"].append("next_page")
-    page = session_state["current_page"]
-    return session_state["pages"][page], f"{page+1} / {len(session_state['pages'])} ページ"
-
-def prev_page():
-    if session_state["current_page"] > 0:
-        session_state["current_page"] -= 1
-        session_state["actions"].append("prev_page")
-    page = session_state["current_page"]
-    return session_state["pages"][page], f"{page+1} / {len(session_state['pages'])} ページ"
-
-# ✅ 読み終えた or リタイア → 次の教材へ
-def finish(action):
-    finished = datetime.datetime.now().isoformat()
-
-    # ✅ ログ1行を保存
-    log_row = {
-        "user_id": session_state["user_id"],
-        "genre": session_state["genre"],
-        "lexile": session_state["lexile"],
-        "passage_id": session_state["passage_id"],
-        "original_lexile": session_state["original_lexile"],
-        "start_time": session_state["start_time"],
-        "finished_time": finished,
-        "user_action": json.dumps(session_state["actions"], ensure_ascii=False),
-    }
+# --- 次へ ---
+def next_page(pages_json, current_page):
+    pages = json.loads(pages_json)
+    if current_page < len(pages)-1:
+        current_page += 1
+        action_log.append({"action": "next", "time": (datetime.utcnow()+timedelta(hours=9)).isoformat()})
+    return (
+        pages[current_page],
+        f"{current_page+1} / {len(pages)}",
+        json.dumps(pages, ensure_ascii=False),
+        current_page
+    )
 
-    save_log_to_hf(log_row)
-    save_local_csv(log_row)
+# --- 前へ ---
+def prev_page(pages_json, current_page):
+    pages = json.loads(pages_json)
+    if current_page > 0:
+        current_page -= 1
+        action_log.append({"action": "prev", "time": (datetime.utcnow()+timedelta(hours=9)).isoformat()})
+    return (
+        pages[current_page],
+        f"{current_page+1} / {len(pages)}",
+        json.dumps(pages, ensure_ascii=False),
+        current_page
+    )
 
-    # ✅ 次の教材選択
-    if action == "finished":
-        next_df = passages_df[passages_df["genre"] == session_state["genre"]]
-    else:  # retire
-        next_df = passages_df[passages_df["genre"] != session_state["genre"]]
+# --- 読み終えた（同ジャンル） or リタイア（別ジャンル） ---
+def finish_or_retire(pages_json, current_page, pid, orig_lex, start_time, action):
+    pages = json.loads(pages_json)
+    now = (datetime.utcnow()+timedelta(hours=9)).isoformat()
 
-    if len(next_df) == 0:
-        return "教材がありません。終了です。", ""
+    # ✅ ログ保存
+    entry = {
+        "user_id": current_user_id,
+        "genre": current_genre,
+        "lexile_assigned": current_lexile,
+        "passage_id": pid,
+        "original_lexile": orig_lex,
+        "start_time": start_time,
+        "finished_time": now,
+        "actions": json.dumps(action_log, ensure_ascii=False),
+        "result": action
+    }
+    save_log(entry)
 
-    row = next_df.sample(1).iloc[0]
-    pid = row["passage_id"]
-    orig_text = row["text"]
-    orig_lex = row["original_lexile_score"]
+    # ✅ 次教材取得
+    if action == "finished":
+        new_pid, new_text, new_orig_lex = get_new_passage(same_genre=True)
+    else:
+        new_pid, new_text, new_orig_lex = get_new_passage(same_genre=False)
 
-    rewritten = rewrite_to_lexile(orig_text, session_state["lexile"])
-    pages = split_into_pages(rewritten)
+    if new_text is None:
+        return "教材がありません", "", "", 0, "", False, "", 0, "", "", 0
 
-    session_state.update({
-        "passage_id": pid,
-        "original_lexile": orig_lex,
-        "pages": pages,
-        "current_page": 0,
-        "start_time": datetime.datetime.now().isoformat(),
-        "actions": ["first_page_displayed"]
-    })
+    rewritten = rewrite_to_lexile(new_text, current_lexile)
+    new_pages = split_pages(rewritten)
+    new_start = (datetime.utcnow()+timedelta(hours=9)).isoformat()
 
-    return pages[0], f"1 / {len(pages)} ページ"
+    # reset action log
+    action_log.clear()
 
-# ✅ Gradio UI
-genres = [
-    "Literature","Science＆Technology","History","Social Science＆Society",
-    "Arts＆Culture","Religion＆Philosophy","Lifestyle＆Hobbies",
-    "Health＆Medicine","Education＆Reference"
-]
+    return (
+        new_pages[0],
+        f"1 / {len(new_pages)}",
+        json.dumps(new_pages, ensure_ascii=False),
+        0,
+        "",
+        True,
+        new_start,
+        len(new_pages),
+        new_pid,
+        new_orig_lex,
+        current_lexile
+    )
+
+# =========================
+#  UI構築
+# =========================
 
 with gr.Blocks() as demo:
-    gr.Markdown("## Lexile Reading Exercise")
+    gr.Markdown("# 📚 Reading Exercise")
 
-    student_id = gr.Textbox(label="学生番号（必須）")
-    lexile = gr.Number(label="受検者のLexile")
-    genre_select = gr.Dropdown(choices=genres, label="ジャンルを1つ選択")
+    student_id_input = gr.Textbox(label="学生番号", placeholder="例: B123456")
+    lexile_input = gr.Number(label="Lexile（例: 900）")
+    genre_input = gr.Dropdown(choices=genres, label="ジャンル")
 
     start_btn = gr.Button("スタート")
 
-    text_display = gr.Textbox(label="教材", lines=15)
-    page_display = gr.Textbox(label="ページ")
-    debug_out = gr.Textbox(label="デバッグ用（任意）")
+    # 表示領域大きめ
+    text_display = gr.Textbox(label="教材", lines=15, interactive=False)
+    page_display = gr.Textbox(label="進行状況", lines=1, interactive=False)
+
+    # hidden
+    hidden_pages = gr.Textbox(visible=False)
+    hidden_page_index = gr.Number(visible=False)
+    hidden_start_time = gr.Textbox(visible=False)
+    hidden_total_pages = gr.Number(visible=False)
+    hidden_passage_id = gr.Textbox(visible=False)
+    hidden_orig_lex = gr.Textbox(visible=False)
+    hidden_assigned_lex = gr.Textbox(visible=False)
+
+    visible_state = gr.State(False)
+
+    feedback = gr.Textbox(label="メッセージ", interactive=False)
 
     prev_btn = gr.Button("前へ")
     next_btn = gr.Button("次へ")
     finish_btn = gr.Button("読み終えた")
     retire_btn = gr.Button("リタイア")
 
-    start_btn.click(start_test, [student_id, genre_select, lexile], [text_display, page_display, debug_out, finish_btn])
-    next_btn.click(next_page, [], [text_display, page_display])
-    prev_btn.click(prev_page, [], [text_display, page_display])
+    # --- start ---
+    start_btn.click(
+        fn=start_test,
+        inputs=[student_id_input, genre_input, lexile_input],
+        outputs=[
+            text_display, page_display,
+            hidden_pages, hidden_page_index,
+            feedback, visible_state,
+            hidden_start_time, hidden_total_pages,
+            hidden_passage_id, hidden_orig_lex, hidden_assigned_lex
+        ]
+    )
+
+    # --- next ---
+    next_btn.click(
+        fn=next_page,
+        inputs=[hidden_pages, hidden_page_index],
+        outputs=[text_display, page_display, hidden_pages, hidden_page_index]
+    )
 
-    finish_btn.click(lambda: finish("finished"), [], [text_display, page_display])
-    retire_btn.click(lambda: finish("retire"), [], [text_display, page_display])
+    # --- prev ---
+    prev_btn.click(
+        fn=prev_page,
+        inputs=[hidden_pages, hidden_page_index],
+        outputs=[text_display, page_display, hidden_pages, hidden_page_index]
+    )
+
+    # --- 読み終えた ---
+    finish_btn.click(
+        fn=lambda p, i, pid, o, st: finish_or_retire(p, i, pid, o, st, "finished"),
+        inputs=[hidden_pages, hidden_page_index, hidden_passage_id, hidden_orig_lex, hidden_start_time],
+        outputs=[
+            text_display, page_display, hidden_pages, hidden_page_index,
+            feedback, visible_state, hidden_start_time,
+            hidden_total_pages, hidden_passage_id, hidden_orig_lex, hidden_assigned_lex
+        ]
+    )
+
+    # --- リタイア ---
+    retire_btn.click(
+        fn=lambda p, i, pid, o, st: finish_or_retire(p, i, pid, o, st, "retire"),
+        inputs=[hidden_pages, hidden_page_index, hidden_passage_id, hidden_orig_lex, hidden_start_time],
+        outputs=[
+            text_display, page_display, hidden_pages, hidden_page_index,
+            feedback, visible_state, hidden_start_time,
+            hidden_total_pages, hidden_passage_id, hidden_orig_lex, hidden_assigned_lex
+        ]
+    )
+
+    # 表示ON/OFF
+    def toggle(show):
+        v = bool(show)
+        return (
+            gr.update(visible=v), gr.update(visible=v),
+            gr.update(visible=v), gr.update(visible=v),
+            gr.update(visible=v)
+        )
+
+    visible_state.change(
+        fn=toggle,
+        inputs=visible_state,
+        outputs=[text_display, page_display, prev_btn, next_btn, feedback]
+    )
 
 demo.launch()