utils.py

import openai
import json
import requests
import pandas as pd


def ask_datorama(question):

    payload_formation = openai.ChatCompletion.create(
        model="gpt-4",
        messages=[{"role": "system", "content": "Given an input question in any language, respond with the relevant dimensions, metrics, and filters. Be creative but the response must be in the correct JSON format. You can only use dimensions called \"Campaign Name\", \"Campaign Category\",  \"Media Buy Name\", \"Day\", \"Weekday\",\"Month\", \"Creative Name\", \"Media Buy Size\". You can only use measurements called \"Impressions\", \"Clicks\", \"Revenue\", \"CTR\", \"CPM\", \"Media Cost\", \"Conversion CPA\", \"Total Conversions\", \"Conversion Rate\", \"Conversion ROAS\". The dateRange variable can be exclusively one of the following: \"CUSTOM\",\"CUSTOM_ON_GOING\", \"YESTERDAY\", \"WEEK_TO_DATE\", \"BI_WEEK_TO_DATE\", \"MONTH_TO_DATE\", \"QUARTER_TO_DATE\", \"YEAR_TO_DATE\", \"PREV_WEEK\", \"PREV_BI_WEEK\", \"PREV_MONTH\", \"PREV_QUARTER\", \"PREV_YEAR\", \"LAST_WEEK\", \"LAST_BI_WEEK\", \"LAST_MONTH\", \"LAST_3_MONTHS\", \"THIS_WEEK\", \"THIS_BI_WEEK\", \"THIS_MONTH\", \"THIS_QUARTER\", \"THIS_YEAR\".\nIf you use a CUSTOM dateRange you'll need to add startDate & endDate, if you use CUSTOM_ONGOING you'll need to add startDate only."},
                  {"role": "user", "content": "What is the day that had more than 1000 impressions with the most clicks of 2021?"},
                  {"role": "assistant", "content": "{\n  \"workspaceId\": 77283,\n  \"dateRange\": \"CUSTOM\",\n  \"startDate\": \"2021-01-01\",\n  \"endDate\": \"2021-12-31\",\n  \"measurements\": [\n    {\n      \"name\": \"Clicks\"\n    },\n    {\n      \"name\": \"Impressions\"\n    }\n  ],\n  \"dimensions\": [\n    \"Day\"\n  ],\n  \"numberMeasurementFilter\": [\n    {\n      \"fieldName\": \"Impressions\",\n      \"operator\": \"GREATER\",\n      \"val\": 1000\n    }\n  ],\n  \"sortConfig\": [\n    {\n      \"sortBy\": \"Clicks\",\n      \"sortOrder\": \"DESC\"\n    }\n  ],\n  \"topResults\": \"1\",\n  \"groupOthers\": false,\n  \"topPerDimension\": true,\n  \"totalDimensions\": []\n}\n\nInput: \"Cuál es la campaña con el mayor número de impresiones del último mes?\"\nOutput: {\n  \"workspaceId\": 77283,\n  \"dateRange\": \"LAST_MONTH\",\n  \"measurements\": [\n    {\n      \"name\": \"Impressions\"\n    }\n  ],\n  \"dimensions\": [\n    \"Campaign Name\"\n  ],\n  \"sortConfig\": [\n    {\n      \"sortBy\": \"Impressions\",\n      \"sortOrder\": \"DESC\"\n    }\n  ],\n  \"topResults\": \"1\",\n  \"groupOthers\": false,\n  \"topPerDimension\": true,\n  \"totalDimensions\": []\n}."},
                  {"role": "user", "content": question}],
        temperature=0.2
    )
    print(question)
    print(payload_formation.choices[0].message["content"])

    url = "https://api.datorama.com/v1/query"

    payload = payload_formation.choices[0].message["content"]
    headers = {
        'Authorization': 'dato-api-c7637555-2d5c-410c-818f-84bf8125db4c',
        'Content-Type': 'application/json'
    }

    response = requests.request("POST", url, headers=headers, data=payload)
    data = json.loads(response.text)

    print(data)

    columns = [dct for dct in data['queryResponseData']['headers']]
    df = pd.DataFrame(data['queryResponseData']['rows'], columns=columns)
    x = df.to_string(index=False).split('\n')
    vals = ['|'.join(ele.split()) for ele in x]
    vals = '\n'.join(str(item) for item in vals)

    answer = openai.ChatCompletion.create(
        model="gpt-4",
        messages=[{"role": "system", "content": "Answer the question in long form for any given language using the pipe-separated dataframe."},
                  {"role": "user", "content": "What was the campaign that generated the most impressions during last month?\nCampaign Name|Impressions\nCampaignA|12345"},
                  {"role": "assistant", "content": "The campaign that generated the most impressions during last month was Campaign A with 12345 impressions."},
                  {"role": "user", "content": question+"\n"+vals}],
        temperature=0.4
    )

    return answer.choices[0].message["content"]