rtapella
commented
1 year ago Example from Jack L : Generally you'd use to model to predict the value based on the input features in cases where you don't have the corresponding answer
So if you trained a model to predict temperature based on pressure and altitude you'd need all 3 to train. Creating that mapping between altitude/pressure and temperature. Then when you only have altitude/pressure you could use it to predict temperature
So that data you'd want to use it with might come from a separate file
and you'd need to track the model as a different category of object likely
So it introduces some new UI/UX elements we never had
Generally the label (in this example temperature) has a higher cost to collect, and there is less of it
So in this example maybe you have a weather station which, for whatever reason can't collect temperature directly. So you stand out there for a day and collect temperature on a different temporary device and then use that with the altitude/pressure collected on your permanent device. Then you train that model which relates pressure/altitude to your day of hand collected temperature values
Then once you train that model you have it for future days
Then the next day you load in your file of altitude/pressure from your weather station and use it with your mode
You didn't hand collect temperatures that day, but you should be able to predict them with the model which takes in altitude/pressure you trained on the day where you had temperatures to correlate with
Make a way to export a regression model and/or use it within CODEX to predict missing data in a column/Feature