Add Week 8 exercise for IbrahimSheriff: SpecialistAgent, FrontierAgen…

week8/community_contributions/IbrahimSheriff/week8_exercise.ipynb

@@ -0,0 +1,348 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# Week 8 Exercise: The Price is Right — IbrahimSheriff\n",
+    "\n",
+    "## Overview\n",
+    "This notebook implements a complete Week 8–style pipeline:\n",
+    "- **SpecialistAgent**: Fine-tuned LLM deployed on Modal\n",
+    "- **FrontierAgent**: RAG + frontier model (GPT) with ChromaDB\n",
+    "- **ChromaDB**: Vector store of product embeddings (SentenceTransformer)\n",
+    "- **Ensemble**: Combined predictor (Specialist + Frontier)\n",
+    "- **Evaluation**: Official `evaluate()` on test set (average absolute error)\n",
+    "- **DealAgentFramework + Gradio UI**: Autonomous deal-hunting and table UI\n",
+    "\n",
+    "## Prerequisites\n",
+    "- Modal token set; HuggingFace secret in Modal (`huggingface-secret`)\n",
+    "- Deploy pricer: from **week8** folder run `uv run modal deploy -m pricer_service`\n",
+    "- `.env`: `HF_TOKEN`, `OPENAI_API_KEY` (for FrontierAgent)\n",
+    "---"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## 1. Setup: path and environment"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import os\n",
+    "import sys\n",
+    "import logging\n",
+    "from pathlib import Path\n",
+    "from dotenv import load_dotenv\n",
+    "\n",
+    "load_dotenv(override=True)\n",
+    "os.environ.setdefault(\"PYTHONIOENCODING\", \"utf-8\")\n",
+    "\n",
+    "notebook_dir = Path.cwd()\n",
+    "week8_root = notebook_dir.parent.parent\n",
+    "if str(week8_root) not in sys.path:\n",
+    "    sys.path.insert(0, str(week8_root))\n",
+    "\n",
+    "logging.getLogger().setLevel(logging.INFO)\n",
+    "print(f\"Week 8 root: {week8_root}\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "required = [\"HF_TOKEN\", \"OPENAI_API_KEY\"]\n",
+    "for var in required:\n",
+    "    status = \"SET\" if os.getenv(var) else \"MISSING\"\n",
+    "    print(f\"  {var}: {status}\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## 2. Load data (Item.from_hub)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from agents.items import Item\n",
+    "from huggingface_hub import login\n",
+    "\n",
+    "hf_token = os.environ.get(\"HF_TOKEN\")\n",
+    "if hf_token:\n",
+    "    login(token=hf_token, add_to_git_credential=False)\n",
+    "\n",
+    "LITE_MODE = False\n",
+    "username = \"ed-donner\"\n",
+    "dataset = f\"{username}/items_lite\" if LITE_MODE else f\"{username}/items_full\"\n",
+    "\n",
+    "train, val, test = Item.from_hub(dataset)\n",
+    "print(f\"Loaded {len(train):,} train, {len(val):,} val, {len(test):,} test items\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## 3. ChromaDB vector store\n",
+    "\n",
+    "We build the vector store in the current folder so DealAgentFramework can use it later. Use a subset for speed if you prefer."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import chromadb\n",
+    "from sentence_transformers import SentenceTransformer\n",
+    "from tqdm.notebook import tqdm\n",
+    "\n",
+    "DB = \"products_vectorstore\"\n",
+    "client = chromadb.PersistentClient(path=DB)\n",
+    "collection = client.get_or_create_collection(\"products\")\n",
+    "\n",
+    "def description(item):\n",
+    "    return (item.summary or item.title or \"\").strip() or getattr(item, \"prompt\", \"\")[:500]\n",
+    "\n",
+    "N_DOCS = 20_000\n",
+    "existing = collection.count()\n",
+    "print(f\"Collection has {existing} documents.\")\n",
+    "if existing == 0:\n",
+    "    print(f\"Populating with {N_DOCS} items...\")\n",
+    "    encoder = SentenceTransformer(\"sentence-transformers/all-MiniLM-L6-v2\")\n",
+    "    for i in tqdm(range(0, min(N_DOCS, len(train)), 1000)):\n",
+    "        batch = train[i : i + 1000]\n",
+    "        docs = [description(it) for it in batch]\n",
+    "        vecs = encoder.encode(docs).astype(float).tolist()\n",
+    "        metas = [{\"category\": getattr(it, \"category\", \"\"), \"price\": it.price} for it in batch]\n",
+    "        ids = [f\"doc_{j}\" for j in range(i, i + len(batch))]\n",
+    "        collection.add(ids=ids, documents=docs, embeddings=vecs, metadatas=metas)\n",
+    "    print(f\"Done. Collection count: {collection.count()}\")\n",
+    "else:\n",
+    "    print(\"Using existing vector store.\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## 4. SpecialistAgent (Modal fine-tuned pricer)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from agents.specialist_agent import SpecialistAgent\n",
+    "\n",
+    "specialist = SpecialistAgent()\n",
+    "sample = test[0]\n",
+    "est = specialist.price(description(sample))\n",
+    "print(f\"Sample: {sample.title[:50]}...\")\n",
+    "print(f\"Actual: ${sample.price:.2f}, Specialist: ${est:.2f}\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## 5. FrontierAgent (RAG + frontier model)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from agents.frontier_agent import FrontierAgent\n",
+    "\n",
+    "frontier = FrontierAgent(collection)\n",
+    "est_f = frontier.price(description(sample))\n",
+    "print(f\"Frontier estimate: ${est_f:.2f}\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## 6. Predictors for evaluator\n",
+    "\n",
+    "Each takes an `Item` and returns a price (for `evaluate(predictor, test)`)."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "def specialist_predictor(item):\n",
+    "    return specialist.price(description(item))\n",
+    "\n",
+    "def frontier_predictor(item):\n",
+    "    return frontier.price(description(item))\n",
+    "\n",
+    "def ensemble_predictor(item):\n",
+    "    s = specialist.price(description(item))\n",
+    "    f = frontier.price(description(item))\n",
+    "    return 0.5 * s + 0.5 * f"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## 7. Run evaluation (Week 8 implementation)\n",
+    "\n",
+    "Uses the same `evaluate()` as week8 day2: 200 items, report and charts. This **determines your result**."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from agents.evaluator import evaluate\n",
+    "\n",
+    "print(\"=== SpecialistAgent ===\")\n",
+    "evaluate(specialist_predictor, test)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "print(\"=== FrontierAgent ===\")\n",
+    "evaluate(frontier_predictor, test)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "print(\"=== Ensemble (Specialist + Frontier) ===\")\n",
+    "evaluate(ensemble_predictor, test)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## 8. Optional: quick run with fewer items\n",
+    "\n",
+    "Uncomment to test with 50 items and 3 workers."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# evaluate(specialist_predictor, test, size=50, workers=3)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## 9. DealAgentFramework + Gradio UI\n",
+    "\n",
+    "Uses the same framework as week8 day5: planning agent, scanner, messenger. The UI shows a table of deals and a button to run one planning cycle. Ensure `products_vectorstore` is in the current directory (we built it above)."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import gradio as gr\n",
+    "from deal_agent_framework import DealAgentFramework\n",
+    "from agents.deals import Opportunity, Deal\n",
+    "\n",
+    "agent_framework = DealAgentFramework()\n",
+    "agent_framework.init_agents_as_needed()\n",
+    "print(\"DealAgentFramework ready.\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "def get_table(opps):\n",
+    "    if not opps:\n",
+    "        return [[\"No deals yet\", \"—\", \"—\", \"—\", \"—\"]]\n",
+    "    return [\n",
+    "        [\n",
+    "            opp.deal.product_description[:80] + (\".\" if len(opp.deal.product_description) > 80 else \"\"),\n",
+    "            f\"${opp.deal.price:.2f}\",\n",
+    "            f\"${opp.estimate:.2f}\",\n",
+    "            f\"${opp.discount:.2f}\",\n",
+    "            opp.deal.url or \"—\",\n",
+    "        ]\n",
+    "        for opp in opps\n",
+    "    ]\n",
+    "\n",
+    "def run_one_cycle():\n",
+    "    agent_framework.run()\n",
+    "    return get_table(agent_framework.memory)\n",
+    "\n",
+    "with gr.Blocks(title=\"The Price is Right\", fill_width=True) as ui:\n",
+    "    gr.Markdown(\"<div style='text-align: center; font-size: 24px'>The Price is Right — IbrahimSheriff</div>\")\n",
+    "    gr.Markdown(\"Deals surfaced by the autonomous agent (Specialist + Frontier + Scanner + Planner).\")\n",
+    "    with gr.Row():\n",
+    "        run_btn = gr.Button(\"Run one planning cycle\")\n",
+    "    tbl = gr.Dataframe(\n",
+    "        headers=[\"Description\", \"Price\", \"Estimate\", \"Discount\", \"URL\"],\n",
+    "        wrap=True,\n",
+    "        row_count=10,\n",
+    "        col_count=5,\n",
+    "        max_height=400,\n",
+    "    )\n",
+    "    run_btn.click(fn=run_one_cycle, outputs=[tbl])\n",
+    "    ui.load(fn=lambda: get_table(agent_framework.memory), outputs=[tbl])\n",
+    "\n",
+    "ui.launch(inbrowser=True)"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "name": "python",
+   "version": "3.11.0"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 4
+}