reflexion-human-eval/hotpotqa_runs/notebooks/CotQA_no_context.ipynb

{
 "cells": [
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Notebook for running Chain-of-Thought with no supporting context experiments"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "import sys, os\n",
    "sys.path.append('..')\n",
    "root = '../root/'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "from util import summarize_trial, log_trial, save_agents\n",
    "import joblib\n",
    "from agents import CoTAgent, ReflexionStrategy"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Load the HotPotQA Sample"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "hotpot = joblib.load('../data/hotpot-qa-distractor-sample.joblib').reset_index(drop = True)"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Define the Reflexion Strategy"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "    NONE: No reflection\n",
      "    LAST_ATTEMPT: Use last reasoning trace in context \n",
      "    REFLEXION: Apply reflexion to the next reasoning trace \n",
      "    LAST_ATTEMPT_AND_REFLEXION: Use last reasoning trace in context and apply reflexion to the next reasoning trace \n",
      "    \n"
     ]
    }
   ],
   "source": [
    "print(ReflexionStrategy.__doc__)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "strategy: ReflexionStrategy = ReflexionStrategy.REFLEXION"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Initialize a CoTAgent for each question"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "from prompts import cot_simple_reflect_agent_prompt, cot_simple_reflect_prompt, cot_simple_agent_prompt\n",
    "from fewshots import COTQA_SIMPLE6, COT_SIMPLE_REFLECTION\n",
    "\n",
    "agents = [CoTAgent(question = row['question'],\n",
    "                   context = '',\n",
    "                   key = row['answer'],\n",
    "                   agent_prompt=cot_simple_agent_prompt if strategy == ReflexionStrategy.NONE else cot_simple_reflect_agent_prompt,\n",
    "                   cot_examples = COTQA_SIMPLE6,\n",
    "                   reflect_prompt = cot_simple_reflect_prompt,\n",
    "                   reflect_examples = COT_SIMPLE_REFLECTION,\n",
    "                      ) for _, row in hotpot.iterrows()]"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Run `n` trials"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "n = 5\n",
    "trial = 0\n",
    "log = ''\n",
    "for i in range(n):\n",
    "    for agent in [a for a in agents if not a.is_correct()]:\n",
    "        agent.run(reflexion_strategy = strategy)\n",
    "        print(f'Answer: {agent.key}')\n",
    "    trial += 1\n",
    "    log += log_trial(agents, trial)\n",
    "    correct, incorrect = summarize_trial(agents)\n",
    "    print(f'Finished Trial {trial}, Correct: {len(correct)}, Incorrect: {len(incorrect)}')"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Save the result log"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "with open(os.path.join(root, 'CoT', 'no_context', strategy.value, f'{len(agents)}_questions_{trial}_trials.txt'), 'w') as f:\n",
    "    f.write(log)\n",
    "save_agents(agents, os.path.join(root, 'CoT', 'no_context', strategy.value, 'agents'))"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "env",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.9.16"
  },
  "orig_nbformat": 4,
  "vscode": {
   "interpreter": {
    "hash": "e23f799cbd2581634725fbf6ce3480ae26192d78438dfafc8efe944acd6490d5"
   }
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}