Update app.py

app.py

@@ -1,187 +1,206 @@
-import os
-import time, random, json, datetime
-import pandas as pd
 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 load_dataset, Dataset, DatasetDict
 
 API_KEY = os.getenv("API_KEY")
-client = OpenAI(
-    base_url="https://openrouter.ai/api/v1",
-    api_key=API_KEY
-)
+HF_TOKEN = os.getenv("HF_TOKEN")
+HF_DATASET = "Toya0421/reading_exercise_logging"
 
-DATASET_REPO = "Toya0421/reading_exercise_logging"
-
-passages_df = pd.read_csv("passage.csv")
+client = OpenAI(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"])
 
+# ✅ 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()
-    return [" ".join(words[i:i+words_per_page]) for i in range(0, len(words), words_per_page)]
-
+    pages = [" ".join(words[i:i+words_per_page]) for i in range(0, len(words), words_per_page)]
+    return pages
 
-def rewrite_to_lexile(text, lexile_score):
+# ✅ 書き換え
+def rewrite_to_lexile(text, target_lexile):
     prompt = f"""
-Rewrite this text to match about {lexile_score}L. Keep meaning. Do not shorten content.
+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.
+
 {text}
 """
-    for _ in range(3):
-        try:
-            res = client.chat.completions.create(
-                model="google/gemma-3-27b-it:free",
-                messages=[{"role": "user", "content": prompt}]
-            )
-            return res.choices[0].message.content.strip()
-        except:
-            time.sleep(3)
-    return text
-
+    resp = client.chat.completions.create(
+        model="gpt-5",
+        messages=[{"role": "user", "content": prompt}],
+        temperature=0.3,
+    )
+    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)
+    else:
+        df.to_csv("full_log.csv", index=False)
+
+# ✅ 状態管理
 session_state = {
     "user_id": None,
-    "lexile": None,
     "genre": None,
-    "pages": [],
-    "page": 0,
+    "lexile": None,
     "passage_id": None,
+    "original_lexile": None,
+    "pages": [],
+    "current_page": 0,
     "actions": [],
-    "first_page_time": None
+    "start_time": None,
 }
 
-used_passages = set()
-
-
-def save_log(action_type):
-    finished = datetime.datetime.now().isoformat()
-    row = {
-        "user_id": session_state["user_id"],
-        "lexile_score": session_state["lexile"],
-        "genre": session_state["genre"],
-        "passage_id": session_state["passage_id"],
-        "first_page_time": session_state["first_page_time"],
-        "finished_time": finished,
-        "finish_type": action_type,
-        "actions": json.dumps(session_state["actions"], ensure_ascii=False)
-    }
-
-    try:
-        ds = load_dataset(DATASET_REPO)
-        df = ds["train"].to_pandas()
-        df = pd.concat([df, pd.DataFrame([row])], ignore_index=True)
-        new_ds = Dataset.from_pandas(df)
-        DatasetDict({"train": new_ds}).push_to_hub(DATASET_REPO)
-    except:
-        new_ds = Dataset.from_pandas(pd.DataFrame([row]))
-        DatasetDict({"train": new_ds}).push_to_hub(DATASET_REPO)
-
-
-def load_new_passage():
-    df = passages_df[passages_df["genre"] == session_state["genre"]]
-    df = df[~df["passage_id"].isin(used_passages)]
+# ✅ テスト開始
+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)
 
-    if df.empty:
-        return "✅ このジャンルの教材は全て終了しました", "", gr.update(interactive=False)
+    session_state["user_id"] = student_id
+    session_state["genre"] = genre
+    session_state["lexile"] = int(lexile)
 
-    row = df.sample(1).iloc[0]
-    used_passages.add(row["passage_id"])
+    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"]
 
-    rewritten = rewrite_to_lexile(row["text"], session_state["lexile"])
+    rewritten = rewrite_to_lexile(orig_text, session_state["lexile"])
     pages = split_into_pages(rewritten)
 
     session_state.update({
-        "passage_id": row["passage_id"],
+        "passage_id": pid,
+        "original_lexile": orig_lex,
         "pages": pages,
-        "page": 0,
-        "first_page_time": datetime.datetime.now().isoformat(),
-        "actions": ["open"]
+        "current_page": 0,
+        "start_time": datetime.datetime.now().isoformat(),
+        "actions": ["first_page_displayed"]
     })
 
-    finish_enabled = (len(pages) == 1)
-    return pages[0], f"1 / {len(pages)}ページ", gr.update(interactive=finish_enabled)
-
-
-def start(user_id, lexile, genre):
-    if not user_id or not genre:
-        return "⚠️学生番号とジャンルを入力してください", "", gr.update(interactive=False)
-
-    session_state["user_id"] = user_id
-    session_state["lexile"] = lexile
-    session_state["genre"] = genre
-
-    return load_new_passage()
-
-
-def next_page():
-    session_state["actions"].append("next")
-
-    if session_state["page"] < len(session_state["pages"]) - 1:
-        session_state["page"] += 1
-
-    p = session_state["page"]
-    is_last = (p == len(session_state["pages"]) - 1)
-
     return (
-        session_state["pages"][p],
-        f"{p+1} / {len(session_state['pages'])}ページ",
-        gr.update(interactive=is_last)
+        pages[0],
+        f"1 / {len(pages)} ページ",
+        "",
+        gr.update(interactive=True)
     )
 
+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():
-    session_state["actions"].append("prev")
+    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),
+    }
 
-    if session_state["page"] > 0:
-        session_state["page"] -= 1
+    save_log_to_hf(log_row)
+    save_local_csv(log_row)
 
-    p = session_state["page"]
-    is_last = (p == len(session_state["pages"]) - 1)
+    # ✅ 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"]]
 
-    return (
-        session_state["pages"][p],
-        f"{p+1} / {len(session_state['pages'])}ページ",
-        gr.update(interactive=is_last)
-    )
+    if len(next_df) == 0:
+        return "教材がありません。終了です。", ""
 
+    row = next_df.sample(1).iloc[0]
+    pid = row["passage_id"]
+    orig_text = row["text"]
+    orig_lex = row["original_lexile_score"]
 
-def finish_reading():
-    session_state["actions"].append("finish")
-    save_log("finish")
-    return load_new_passage()
+    rewritten = rewrite_to_lexile(orig_text, session_state["lexile"])
+    pages = split_into_pages(rewritten)
 
+    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"]
+    })
 
-def retire():
-    session_state["actions"].append("retire")
-    save_log("retire")
-    return load_new_passage()
+    return pages[0], f"1 / {len(pages)} ページ"
 
+# ✅ Gradio UI
+genres = [
+    "Literature","Science＆Technology","History","Social Science＆Society",
+    "Arts＆Culture","Religion＆Philosophy","Lifestyle＆Hobbies",
+    "Health＆Medicine","Education＆Reference"
+]
 
-# ==== UI ====
 with gr.Blocks() as demo:
-    gr.Markdown("### 📚 Reading Exercise")
+    gr.Markdown("## Lexile Reading Exercise")
 
-    user_id = gr.Textbox(label="学生番号（必須）")
-    lexile = gr.Number(label="Lexile")
-    genre_select = gr.Dropdown(
-        label="ジャンルを選択（1つ）",
-        choices=sorted(passages_df["genre"].unique())
-    )
-    start_btn = gr.Button("スタート")
+    student_id = gr.Textbox(label="学生番号（必須）")
+    lexile = gr.Number(label="��検者のLexile")
+    genre_select = gr.Dropdown(choices=genres, label="ジャンルを1つ選択")
 
-    text_display = gr.Textbox(label="教材", lines=25)
-    status = gr.Textbox(label="ページ")
+    start_btn = gr.Button("スタート")
 
-    with gr.Row():
-        prev_btn = gr.Button("前へ")
-        next_btn = gr.Button("次へ")
+    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）")
 
-    finish_btn = gr.Button("読み終えた", 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, [user_id, lexile, genre_select], [text_display, status, finish_btn])
-    next_btn.click(next_page, outputs=[text_display, status, finish_btn])
-    prev_btn.click(prev_page, outputs=[text_display, status, finish_btn])
-    finish_btn.click(finish_reading, outputs=[text_display, status, finish_btn])
-    retire_btn.click(retire, outputs=[text_display, status, finish_btn])
+    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])
+
+    finish_btn.click(lambda: finish("finished"), [], [text_display, page_display])
+    retire_btn.click(lambda: finish("retire"), [], [text_display, page_display])
 
 demo.launch()