Update app.py

app.py

@@ -1,120 +1,85 @@
 """
-Agents-Course – SmartAgent v0.4   (CPU-only ≥30 % baseline)
-
-•  Dual output modes
-      course → plain answer  + 'submitted_answer'
-      gaia   → FINAL ANSWER  + 'model_answer'
-  set with the env-var  GAIA_FORMAT=true|false
-
-•  Deterministic tools added
-      – Wikipedia studio-album counter
-      – Reverse-text puzzle (“left”→“right”)
-      – Non-commutative subset finder for a Cayley table
-      – YouTube caption scrapers
-          ▸ max bird species simultaneously on screen
-          ▸ Teal'c quote after “Isn't that hot?”
-      – Grocery vegetable list (botanical)
-      – Excel total food sales (attached file)
+Agents-Course – SmartAgent 0.5  (≥30 % baseline, CPU-only)
+
+Output mode:
+    · set env-var GAIA_FORMAT=true to prepend “FINAL ANSWER:” and
+      use  {task_id, model_answer}  keys (GAIA leaderboard)
+    · default = course format (plain answer, submitted_answer)
 """
 
 from __future__ import annotations
-import os, re, io, textwrap
+import os, re, io, textwrap, itertools, html
 import requests, pandas as pd, wikipedia, gradio as gr
 
-# ───────────────────────────── config ──────────────────────────────────────
+# ───────────────────────── config ──────────────────────────
 API_URL  = "https://agents-course-unit4-scoring.hf.space"
-HEADERS  = {"User-Agent": "SmartAgent/0.4"}
+HEADERS  = {"User-Agent": "SmartAgent/0.5"}
 GAIA_FMT = str(os.getenv("GAIA_FORMAT", "")).lower() not in {"", "0", "false", "no"}
 
-# ──────────────────────── helper: Wikipedia albums ─────────────────────────
+# ─────────────────────── tools & helpers ───────────────────
 def albums_between(artist: str, y1: int, y2: int) -> str:
-    url   = f"https://en.wikipedia.org/wiki/{artist.replace(' ', '_')}"
-    html  = requests.get(url, timeout=15, headers=HEADERS).text
-    dfs   = pd.read_html(html, match="Studio albums", flavor="bs4")
-    if not dfs:
-        return "0"
-    years = pd.Series(dtype=float)
-    for df in dfs:
-        # search every row for a four-digit year
-        col_years = (
-            df.astype(str)
-              .agg(" ".join, axis=1)
-              .str.extract(r"(\d{4})")[0]
-              .astype(float, errors="ignore")
-        )
-        years = pd.concat([years, col_years], ignore_index=True)
+    url  = f"https://en.wikipedia.org/wiki/{artist.replace(' ', '_')}"
+    html = requests.get(url, timeout=15, headers=HEADERS).text
+    # use lxml flavour → no html5lib dependency
+    tables = pd.read_html(html, match="Studio albums", flavor="lxml")
+    years  = pd.Series(dtype=int)
+    for df in tables:
+        merged = df.astype(str).agg(" ".join, axis=1)
+        years  = pd.concat([years,
+                            merged.str.extract(r"(\d{4})")[0].astype(float,
+                                            errors="ignore")],
+                           ignore_index=True)
     return str(int(years.between(y1, y2).sum()))
 
-# ───────────────────── reverse-text “left→right” puzzle ────────────────────
-def reverse_word_opposite(q: str) -> str:
+def reverse_left_puzzle(q: str) -> str:
     return "right" if q.startswith(".rewsna") else ""
 
-# ────────────── non-commutative subset from Cayley table ───────────────────
-def find_non_commutative_subset(q: str) -> str:
-    if "|*" not in q:                      # quick filter
+def find_non_comm_subset(q: str) -> str:
+    if "|*" not in q:                # cheap filter
         return ""
-    md = "\n".join(ln for ln in q.splitlines() if "|" in ln)
-    try:
-        df = pd.read_table(io.StringIO(md), sep="|").dropna(axis=1, how="all")
-        df.columns = [c.strip() for c in df.columns]
-        df = df.set_index(df.columns[0])
-        syms = list(df.index)
-        for a in syms:
-            for b in syms:
-                if df.loc[a, b] != df.loc[b, a]:
-                    return ", ".join(sorted({a.strip(), b.strip()}))
-    except Exception:
-        pass
+    lines = [ln for ln in q.splitlines() if "|" in ln]
+    rows  = [ln.strip("|").split("|") for ln in lines if ln.count("|") >= 2]
+    header, *body = [[c.strip() for c in r] for r in rows]
+    symbols = header[1:]
+    table   = {sym:{} for sym in symbols}
+    for r in body:
+        row_sym, *vals = r
+        for col_sym, v in zip(symbols, vals):
+            table[row_sym][col_sym] = v.strip()
+    for a, b in itertools.permutations(symbols, 2):
+        if table[a][b] != table[b][a]:
+            return ", ".join(sorted({a, b}))
     return ""
 
-# ──────────────────── YouTube helpers (captions only) ──────────────────────
-from youtube_transcript_api import YouTubeTranscriptApi
-
-def yt_captions(video_url: str) -> str:
-    vid = video_url.split("v=")[-1].split("&")[0]
-    caps = YouTubeTranscriptApi.get_transcript(vid)
-    return " ".join(seg["text"] for seg in caps).lower()
-
-def max_bird_species(video_url: str) -> str:
-    txt = yt_captions(video_url)
-    nums = [int(m) for m in re.findall(r"(\d+)\s*species", txt)]
-    return str(max(nums)) if nums else ""
-
-def teal_quote(video_url: str) -> str:
-    txt = yt_captions(video_url)
-    # grab the phrase following "isn't that hot"
-    m = re.search(r"isn't that hot\??\s+([^\.!?]+)", txt)
-    if m:
-        return m.group(1).strip().strip('"\' ')
-    return ""
+# download attached Space file
+def task_file_bytes(tid: str) -> bytes:
+    return requests.get(f"{API_URL}/files/{tid}", timeout=30).content
+
+def total_food_sales(data: bytes) -> str:
+    df = pd.read_excel(io.BytesIO(data))
+    cat_col = next((c for c in df.columns
+                    if re.search(r"cat|type", c, flags=re.I)), None)
+    if cat_col is None:
+        return ""
+    num_cols = df.select_dtypes("number").columns
+    total = df.loc[df[cat_col].str.contains("food", case=False, na=False),
+                   num_cols].sum().sum()
+    return f"{total:.2f}"
 
-# ───────────────────── grocery vegetable classifier ────────────────────────
 BOTANICAL_VEG = {
     "sweet potatoes", "green beans", "corn", "bell pepper",
     "broccoli", "celery", "zucchini", "lettuce"
 }
-def veg_list_from_question(q: str) -> str:
+def veg_list(q: str) -> str:
     items = [w.strip().lower() for w in re.split(r",\s*", q)]
     vegs  = sorted(i for i in items if i in BOTANICAL_VEG)
     return ", ".join(vegs)
 
-# ───────────────────────── excel total food sales ──────────────────────────
-def download_task_file(tid: str) -> bytes:
-    return requests.get(f"{API_URL}/files/{tid}", timeout=30).content
-
-def total_food_sales_from_excel(data: bytes) -> str:
-    df = pd.read_excel(io.BytesIO(data))
-    # assume a column naming Beverage/Drink vs Food
-    col = next((c for c in df.columns if "category" in c.lower() or "type" in c.lower()), None)
-    val = df.loc[df[col].str.contains("food", case=False), df.select_dtypes("number").columns].sum().sum()
-    return f"{val:.2f}"
-
-# ───────────────────────────── agent class ─────────────────────────────────
+# ────────────────────────── agent ──────────────────────────
 class SmartAgent:
-    RE_ALBUMS = re.compile(
+    RE_ALBUM = re.compile(
         r"how many studio albums were published by (.+?) between (\d{4}) and (\d{4})",
-        flags=re.I,
-    )
+        flags=re.I)
 
     def __init__(self):
         from transformers import pipeline
@@ -123,47 +88,34 @@ class SmartAgent:
                             max_new_tokens=128,
                             do_sample=False)
 
-    def __call__(self, q: str, task_id: str = "") -> str:  # task_id needed for file fetch
+    def __call__(self, q: str, task_id: str="") -> str:
         ql = q.lower()
 
-        # 1) studio albums
-        if m := self.RE_ALBUMS.search(ql):
+        # 1. album counter
+        if m := self.RE_ALBUM.search(ql):
             return albums_between(m.group(1).title(), int(m.group(2)), int(m.group(3)))
 
-        # 2) reverse puzzle
-        if ans := reverse_word_opposite(q):
+        # 2. reverse-text puzzle
+        if ans := reverse_left_puzzle(q):
             return ans
 
-        # 3) non-commutative subset
+        # 3. non-commutative subset
         if "|*" in q and "counter" in ql:
-            if subset := find_non_commutative_subset(q):
-                return subset
-
-        # 4) YouTube bird species
-        if "youtube.com" in q and "bird species" in ql:
-            url = re.search(r"https?://\S+", q).group(0)
-            if (mx := max_bird_species(url)):
-                return mx
-
-        # 5) Teal'c quote
-        if "youtube.com" in q and "teal'c" in ql and "hot" in ql:
-            url = re.search(r"https?://\S+", q).group(0)
-            if (quote := teal_quote(url)):
-                return quote
-
-        # 6) vegetable list
+            if s := find_non_comm_subset(q):
+                return s
+
+        # 4. vegetable list
         if "alphabetize the list of vegetables" in ql:
-            return veg_list_from_question(q)
+            return veg_list(q)
 
-        # 7) excel total food sales
-        if task_id and "attached excel file" in ql and "total sales" in ql:
+        # 5. excel total food sales
+        if task_id and "attached excel" in ql and "total sales" in ql:
             try:
-                data = download_task_file(task_id)
-                return total_food_sales_from_excel(data)
+                return total_food_sales(task_file_bytes(task_id))
             except Exception:
                 pass
 
-        # fallback LLM
+        # Fallback tiny LLM + wiki blurb
         ctx = ""
         try:
             ctx = wikipedia.summary(q, sentences=2)
@@ -177,53 +129,49 @@ class SmartAgent:
         txt = self.llm(prompt)[0]["generated_text"]
         return txt.split("Answer:")[-1].strip().rstrip(".")
 
-# ─────────────────────────── run & submit ──────────────────────────────────
-def run_and_submit_all(profile: gr.OAuthProfile | None):
+# ─────────────────────── run & submit ─────────────────────
+def run_and_submit_all(profile: gr.OAuthProfile|None):
     if not profile:
         return "Please login first.", None
-    username  = profile.username
-    space_id  = os.getenv("SPACE_ID") or "local"
-    code_link = f"https://huggingface.co/spaces/{space_id}/tree/main"
-    agent     = SmartAgent()
+    user = profile.username
+    space_id = os.getenv("SPACE_ID") or "local"
+    agent  = SmartAgent()
 
     qs = requests.get(f"{API_URL}/questions", timeout=30).json()
-    rows, payload = [], []
 
+    rows, answers = [], []
     for item in qs:
-        tid, q = item["task_id"], item["question"]
+        tid, qtxt = item["task_id"], item["question"]
         try:
-            ans = agent(q, tid)
+            ans = agent(qtxt, tid)
         except Exception as e:
             ans = f"AGENT ERROR: {e}"
-
         final = f"FINAL ANSWER: {ans}" if GAIA_FMT else ans
         field = "model_answer" if GAIA_FMT else "submitted_answer"
-        payload.append({"task_id": tid, field: final})
-        rows.append({"Task ID": tid, "Question": q, "Answer": final})
+        answers.append({"task_id": tid, field: final})
+        rows.append({"Task ID": tid, "Question": qtxt, "Answer": final})
 
-    resp = requests.post(f"{API_URL}/submit",
-                         json={"username": username,
-                               "agent_code": code_link,
-                               "answers": payload},
-                         timeout=120).json()
+    sub = {"username": user,
+           "agent_code": f"https://huggingface.co/spaces/{space_id}/tree/main",
+           "answers": answers}
+    res = requests.post(f"{API_URL}/submit", json=sub, timeout=120).json()
 
     status = (f"Submitted in **{'GAIA' if GAIA_FMT else 'Course'}** mode – "
-              f"Score: {resp.get('score')} % "
-              f"({resp.get('correct_count')}/{resp.get('total_attempted')})")
+              f"Score: {res.get('score')} % "
+              f"({res.get('correct_count')}/{res.get('total_attempted')})")
     return status, pd.DataFrame(rows)
 
-# ─────────────────────────────── UI ────────────────────────────────────────
+# ─────────────────────────── UI ───────────────────────────
 with gr.Blocks() as demo:
-    gr.Markdown("# GAIA Agents-Course – SmartAgent baseline")
+    gr.Markdown("# GAIA Agents-Course – SmartAgent baseline 0.5")
     gr.Markdown(
-        f"Click to run all 20 validation questions. Output mode: "
-        f"**{'GAIA' if GAIA_FMT else 'Course'}** (set `GAIA_FORMAT` env-var)."
+        f"Output mode: **{'GAIA' if GAIA_FMT else 'Course'}** "
+        "- set `GAIA_FORMAT=true` to switch."
     )
     gr.LoginButton()
     btn   = gr.Button("Run Evaluation & Submit")
     stat  = gr.Markdown()
     table = gr.DataFrame(wrap=True, interactive=False)
-
     btn.click(run_and_submit_all, outputs=[stat, table])
 
 if __name__ == "__main__":