Skip to content

Commit

Permalink
Merge pull request #782 from expectedparrot/speedup
Browse files Browse the repository at this point in the history 
Speedup
  • Loading branch information
@apostolosfilippas
apostolosfilippas authored Jul 21, 2024
2 parents 29a18ed + 6c468ae commit 5aaec25
Show file tree
Hide file tree
Showing 152 changed files with 9,835 additions and 3,852 deletions.
39 changes: 23 additions & 16 deletions docs/notebooks/adding_metadata.ipynb
View file
Original file line number Diff line number Diff line change
Expand Up @@ -58,13 +58,13 @@
"outputs": [],
"source": [
"q_reference = QuestionFreeText(\n",
" question_name = \"reference\",\n",
" question_text = \"What is this headline referring to: {{ headline }}\"\n",
" question_name=\"reference\",\n",
" question_text=\"What is this headline referring to: {{ headline }}\",\n",
")\n",
"\n",
"q_frontpage = QuestionYesNo(\n",
" question_name = \"frontpage\",\n",
" question_text = \"Is this story likely to be on the front page of the newspaper: {{ headline }}\"\n",
" question_name=\"frontpage\",\n",
" question_text=\"Is this story likely to be on the front page of the newspaper: {{ headline }}\",\n",
")\n",
"\n",
"survey = Survey([q_reference, q_frontpage])"
Expand Down Expand Up @@ -98,7 +98,7 @@
" \"New York Yankees Win First Pennant in Franchise History\",\n",
" \"Subway Expansion Project Approved by City Council\",\n",
" \"Harlem Renaissance: New Wave of Cultural Expression\",\n",
" \"Mayor Announces New Housing Initiative for Veterans\"\n",
" \"Mayor Announces New Housing Initiative for Veterans\",\n",
" ],\n",
" \"date\": [\n",
" \"1918-11-11\",\n",
Expand All @@ -110,7 +110,7 @@
" \"1918-09-30\",\n",
" \"1918-08-18\",\n",
" \"1918-04-25\",\n",
" \"1918-11-20\"\n",
" \"1918-11-20\",\n",
" ],\n",
" \"author\": [\n",
" \"John Doe\",\n",
Expand All @@ -122,7 +122,7 @@
" \"William Davis\",\n",
" \"Barbara Wilson\",\n",
" \"Charles Miller\",\n",
" \"Elizabeth Taylor\"\n",
" \"Elizabeth Taylor\",\n",
" ],\n",
" \"section\": [\n",
" \"Front Page\",\n",
Expand All @@ -134,8 +134,8 @@
" \"Sports\",\n",
" \"City News\",\n",
" \"Culture\",\n",
" \"Housing\"\n",
" ]\n",
" \"Housing\",\n",
" ],\n",
"}\n",
"\n",
"df = pd.DataFrame(data)"
Expand All @@ -156,12 +156,17 @@
"metadata": {},
"outputs": [],
"source": [
"scenarios = [Scenario({\n",
" \"headline\": row['headline'],\n",
" \"date\": row['date'],\n",
" \"author\": row['author'],\n",
" \"section\": row['section']\n",
"}) for _, row in df.iterrows()]"
"scenarios = [\n",
" Scenario(\n",
" {\n",
" \"headline\": row[\"headline\"],\n",
" \"date\": row[\"date\"],\n",
" \"author\": row[\"author\"],\n",
" \"section\": row[\"section\"],\n",
" }\n",
" )\n",
" for _, row in df.iterrows()\n",
"]"
]
},
{
Expand Down Expand Up @@ -669,7 +674,9 @@
}
],
"source": [
"results.select(\"headline\", \"date\", \"author\", \"section\", \"reference\", \"frontpage\").print(format=\"rich\")"
"results.select(\"headline\", \"date\", \"author\", \"section\", \"reference\", \"frontpage\").print(\n",
" format=\"rich\"\n",
")"
]
}
],
Expand Down
167 changes: 101 additions & 66 deletions docs/notebooks/agentifying_responses.ipynb
View file
Original file line number Diff line number Diff line change
Expand Up @@ -29,7 +29,7 @@
"source": [
"import pandas as pd\n",
"\n",
"df = pd.read_csv('newsletter_survey_responses.csv')"
"df = pd.read_csv(\"newsletter_survey_responses.csv\")"
]
},
{
Expand Down Expand Up @@ -121,7 +121,7 @@
}
],
"source": [
"df[['question_type', 'question_text', 'question_options']].drop_duplicates()"
"df[[\"question_type\", \"question_text\", \"question_options\"]].drop_duplicates()"
]
},
{
Expand Down Expand Up @@ -236,7 +236,7 @@
}
],
"source": [
"df[['question_text', 'response']].sample(8)"
"df[[\"question_text\", \"response\"]].sample(8)"
]
},
{
Expand Down Expand Up @@ -264,33 +264,33 @@
"outputs": [],
"source": [
"# Change the set of all responses into a dictionary\n",
"all_responses = df.to_dict(orient = 'records')\n",
"all_responses = df.to_dict(orient=\"records\")\n",
"\n",
"# Initialize a dictionary to store responses by respondent_id\n",
"respondents_dict = {}\n",
"\n",
"# Iterate over each row in the responses\n",
"for row in all_responses:\n",
" respondent_id = row['respondent_id']\n",
" question_id = row['question_id']\n",
" question_text = row['question_text']\n",
" question_options = row['question_options']\n",
" response = row['response']\n",
" \n",
" respondent_id = row[\"respondent_id\"]\n",
" question_id = row[\"question_id\"]\n",
" question_text = row[\"question_text\"]\n",
" question_options = row[\"question_options\"]\n",
" response = row[\"response\"]\n",
"\n",
" # Format each response as background information for the relevant respondent\n",
" formatted_string = (\n",
" f\"You were asked: '{question_text}' \"\n",
" f\"The answer options were: '{question_options}'. \"\n",
" f\"You responded: '{response}'\"\n",
" )\n",
" \n",
"\n",
" # Add the information to the respondent's dictionary\n",
" if respondent_id not in respondents_dict:\n",
" respondents_dict[respondent_id] = {}\n",
" \n",
"\n",
" respondents_dict[respondent_id][question_id] = formatted_string\n",
"\n",
"# Print the new dictionary \n",
"# Print the new dictionary\n",
"# print(respondents_dict)"
]
},
Expand All @@ -316,35 +316,43 @@
"from rich import print\n",
"\n",
"# Interview topic\n",
"interview_topic = 'home construction'\n",
"interview_title = 'Home Construction Newsletter Follow-up Interview'\n",
"interview_topic = \"home construction\"\n",
"interview_title = \"Home Construction Newsletter Follow-up Interview\"\n",
"\n",
"# Persona for the interview subjects\n",
"interview_subject_persona = textwrap.dedent(f\"\"\"\\\n",
"interview_subject_persona = textwrap.dedent(\n",
" f\"\"\"\\\n",
"You are a professional with a keen interest in {interview_topic}.\n",
"\"\"\")\n",
"\"\"\"\n",
")\n",
"\n",
"# Instructions for the interview subject agents\n",
"interview_subject_instructions = textwrap.dedent(f\"\"\"\\\n",
"interview_subject_instructions = textwrap.dedent(\n",
" f\"\"\"\\\n",
"You recently completed a reader survey about the newsletter of a \n",
"{interview_topic} expert who also produces a popular blog and podcast. \n",
"Now they are asking you some follow-on questions.\n",
"\"\"\")\n",
"\"\"\"\n",
")\n",
"\n",
"# Persona for the interviewer agent\n",
"interviewer_persona = textwrap.dedent(f\"\"\"\\\n",
"interviewer_persona = textwrap.dedent(\n",
" f\"\"\"\\\n",
"You are an well-known expert on {interview_topic} who produces \n",
"a popular newsletter, blog and podcast on {interview_topic}. \n",
"\"\"\")\n",
"\"\"\"\n",
")\n",
"\n",
"# Instructions for the interviewer agent\n",
"interviewer_instructions = textwrap.dedent(f\"\"\"\\\n",
"interviewer_instructions = textwrap.dedent(\n",
" f\"\"\"\\\n",
"You recently conducted a reader survey about your newsletter on\n",
"{interview_topic}. Now you are asking respondents some follow-on questions.\n",
"\"\"\")\n",
"\"\"\"\n",
")\n",
"\n",
"# Total number of questions to ask in the interview\n",
"total_questions = 5 "
"total_questions = 5"
]
},
{
Expand Down Expand Up @@ -395,18 +403,20 @@
"\n",
"# Iterate over the respondents' data\n",
"for respondent_id, questions in respondents_dict.items():\n",
" \n",
"\n",
" # Initialize the traits for each agent\n",
" traits = {}\n",
" \n",
"\n",
" # Iterate over the questions\n",
" for question_id, formatted_string in questions.items():\n",
" traits[f\"question_id_{question_id}\"] = formatted_string\n",
" \n",
"\n",
" # Create the agent and add it to the agents list\n",
" agent = Agent(name = f\"Respondent {respondent_id}\", \n",
" traits = traits, \n",
" instruction = interview_subject_instructions)\n",
" agent = Agent(\n",
" name=f\"Respondent {respondent_id}\",\n",
" traits=traits,\n",
" instruction=interview_subject_instructions,\n",
" )\n",
" interview_subjects.append(agent)\n",
"\n",
"# Inspecting the first one\n",
Expand All @@ -429,9 +439,10 @@
"metadata": {},
"outputs": [],
"source": [
"# Create an agent for the interviewer \n",
"interviewer = Agent(traits = {'persona':interviewer_persona},\n",
" instruction = interviewer_instructions)"
"# Create an agent for the interviewer\n",
"interviewer = Agent(\n",
" traits={\"persona\": interviewer_persona}, instruction=interviewer_instructions\n",
")"
]
},
{
Expand All @@ -456,77 +467,101 @@
"\n",
"# Selecting a language model to use\n",
"# Model.available() # To see a list of all available models\n",
"model = Model('gpt-4')\n",
"model = Model(\"gpt-4\")\n",
"\n",
"\n",
"def get_next_question(subject, researcher, dialog_so_far):\n",
" scenario = Scenario({'subject': str(subject.traits), 'dialog_so_far': dialog_so_far})\n",
" scenario = Scenario(\n",
" {\"subject\": str(subject.traits), \"dialog_so_far\": dialog_so_far}\n",
" )\n",
" meta_q = QuestionFreeText(\n",
" question_name = 'next_question',\n",
" question_text = \"\"\"\n",
" question_name=\"next_question\",\n",
" question_text=\"\"\"\n",
" This is the background information for the interview subject: {{ subject }}\n",
" This is your current dialog with the interview subject: {{ dialog_so_far }}\n",
" What question you would ask the interview subject next?\n",
" \"\"\"\n",
" \"\"\",\n",
" )\n",
" question_text = (\n",
" meta_q.by(model)\n",
" .by(researcher)\n",
" .by(scenario)\n",
" .run()\n",
" .select(\"next_question\")\n",
" .first()\n",
" )\n",
" question_text = meta_q.by(model).by(researcher).by(scenario).run().select(\"next_question\").first()\n",
" return question_text\n",
"\n",
"\n",
"def get_response_to_question(question_text, subject, dialog_so_far):\n",
" q_to_subject = QuestionFreeText(\n",
" question_name = 'question',\n",
" question_text = f\"\"\"\n",
" question_name=\"question\",\n",
" question_text=f\"\"\"\n",
" This is your current dialog with the interview subject: {dialog_so_far}.\n",
" You are now being asked:\"\"\" + question_text\n",
" You are now being asked:\"\"\"\n",
" + question_text,\n",
" )\n",
" response = q_to_subject.by(model).by(subject).run().select('question').first()\n",
" response = q_to_subject.by(model).by(subject).run().select(\"question\").first()\n",
" return response\n",
"\n",
"\n",
"def ask_question(subject, researcher, dialog_so_far):\n",
" question_text = get_next_question(subject, researcher, dialog_so_far)\n",
" response = get_response_to_question(question_text, subject, dialog_so_far)\n",
"\n",
" print(' \\nQuestion: \\n\\n' + question_text + '\\n\\nResponse: \\n\\n' + response)\n",
" \n",
" return {'question': question_text, 'response': response}\n",
" print(\" \\nQuestion: \\n\\n\" + question_text + \"\\n\\nResponse: \\n\\n\" + response)\n",
"\n",
" return {\"question\": question_text, \"response\": response}\n",
"\n",
"\n",
"def dialog_to_string(d):\n",
" return '\\n'.join([f\"Question: {d['question']}\\nResponse: {d['response']}\" for d in d])\n",
" return \"\\n\".join(\n",
" [f\"Question: {d['question']}\\nResponse: {d['response']}\" for d in d]\n",
" )\n",
"\n",
"\n",
"def clean_dict(d):\n",
" \"\"\"Convert dictionary to string and remove braces.\"\"\"\n",
" return str(d).replace('{', '').replace('}', '')\n",
" return str(d).replace(\"{\", \"\").replace(\"}\", \"\")\n",
"\n",
"\n",
"def summarize_interview(subject, interview_topic, dialog_so_far, researcher):\n",
" interview_subject_name = subject['name']\n",
" interview_subject_traits = subject['traits']\n",
" interview_subject_name = subject[\"name\"]\n",
" interview_subject_traits = subject[\"traits\"]\n",
" summary_q = QuestionFreeText(\n",
" question_name = 'summary',\n",
" question_text = (\n",
" f\"You have just conducted the following interview of {interview_subject_name} \"\n",
" f\"who has these traits: {clean_dict(interview_subject_traits)} \"\n",
" f\"The topic of the interview was {interview_topic}. \"\n",
" f\"Please draft a summary of the interview: {clean_dict(dialog_so_far)}\")\n",
" question_name=\"summary\",\n",
" question_text=(\n",
" f\"You have just conducted the following interview of {interview_subject_name} \"\n",
" f\"who has these traits: {clean_dict(interview_subject_traits)} \"\n",
" f\"The topic of the interview was {interview_topic}. \"\n",
" f\"Please draft a summary of the interview: {clean_dict(dialog_so_far)}\"\n",
" ),\n",
" )\n",
" themes_q = QuestionFreeText(\n",
" question_name = 'themes',\n",
" question_text = 'List the major themes of the interview.'\n",
" question_name=\"themes\", question_text=\"List the major themes of the interview.\"\n",
" )\n",
" survey = Survey([summary_q, themes_q]).set_full_memory_mode()\n",
" results = survey.by(model).by(researcher).run()\n",
" summary = results.select('summary').first()\n",
" themes = results.select('themes').first()\n",
" print('\\n\\nSummary:\\n\\n' + summary + '\\n\\nThemes:\\n\\n' + themes)\n",
" summary = results.select(\"summary\").first()\n",
" themes = results.select(\"themes\").first()\n",
" print(\"\\n\\nSummary:\\n\\n\" + summary + \"\\n\\nThemes:\\n\\n\" + themes)\n",
"\n",
"\n",
"def conduct_interview(subject, researcher, interview_topic):\n",
"\n",
" print('\\n\\nInterview subject: ' + subject['name'] + '\\n\\nInterview topic: ' + interview_topic)\n",
" \n",
" dialog_so_far = [] \n",
" \n",
" print(\n",
" \"\\n\\nInterview subject: \"\n",
" + subject[\"name\"]\n",
" + \"\\n\\nInterview topic: \"\n",
" + interview_topic\n",
" )\n",
"\n",
" dialog_so_far = []\n",
"\n",
" for i in range(total_questions):\n",
" result = ask_question(subject, researcher, dialog_to_string(dialog_so_far))\n",
" dialog_so_far.append(result)\n",
" \n",
"\n",
" summarize_interview(subject, interview_topic, dialog_so_far, researcher)"
]
},
Expand Down
Loading

0 comments on commit 5aaec25

Please sign in to comment.