0

I have done a python program using the openai.ChatCompletion.create to interface to the ChatGPT API. I am using a messages with two roles: user with the user prompt as content, and assistant loaded with current sensor data as content, which I get from a set of sensors using MQTT. The sensor data is original JSON, but I formatted it like this:

Temperature: 22.07 ºC
Humidity: 62.64 %RH
TVOC: 0 ppb
CO2: 400 ppm
PM1: 0 µg/m3
PM2.5: 0 µg/m3
PM10: 1 µg/m3

My idea is to use a clear language to ask What is the temperature? and What is the dew point? and expect ChatGPT to just give me the first and calculate the second based on the temperature and humidity available in the data.

I don't want to train a model, since the data changes constantly.

Is there a way to feed sensor data to ChatGPT API? (I have also tried with role system, btw)

Gotfredsen
  • 56
  • 4
  • Ok, this might not be the right way, but I got it working by adding the sensor data to my prompt behind the scene, like so: `final_prompt = f'Based on this info: {sensor_data}, can you help with this question: {prompt.capitalize()}'` – Gotfredsen Aug 21 '23 at 15:50

1 Answers1

0

Ok, so I managed to figure it out with some trial and error:

My API call look like this:

        response = openai.ChatCompletion.create(
            model="gpt-3.5-turbo",
            temperature=0.6,
            messages=[
                {
                    "role": "system",
                    "content": "You are a helful assistant that helps with sensor data as provided (temperatures all in ºC, humidity in %RH, light in lux and cloud base in meters). You can assist with calculating other values based on the sensor data. You can put the sensor data into context.",
                },
                {"role": "user", "content": prompt},
                {"role": "assistant", "content": sensor_data},
            ],
        )

The prompt is from HTML input, such as "Is it dry?"

The sensor_data is a text string with the data in a flat format like this:

Temperature: 15.21
Humidity: 82.97
Dew Point: 12.34
Heat Index: 14.96
Wet-Bulb Temperature: 13.27
Cloud Base: 350.38
Cloud Temperature: 11.77
Ambient Light: 30
UV Index: 0.02

The response to the above question looked like this:

Based on the provided data, the humidity is relatively high at 82.97%RH, which suggests that the air contains a significant amount of moisture. However, without additional information, it is difficult to determine if the conditions are considered dry.

The sensor_data variable is updated every 5 seconds from a MQTT server, but that could be from anywhere.

Gotfredsen
  • 56
  • 4