Hey! So far we’ve focused on classification and anomaly detection, but we’re naturally considering other machine learning classifiers, and I’d love to hear from you if you have a regression usecase (predict a scalar value based on your input data). We have an alpha version of regression available in the Studio right now (with full support for DSP blocks and deployment to device), and would love to see if it solves your usecase and how we can improve the experience.
I can see use cases for visually identifying painted lines (e.g. roads) or intersections (e.g. wall-to-floor transition) and fitting a line to those points in attempt to have a robot follow them.
The other is to use a distance sensor to measure the distance from the side or corner of a robot to a wall (or other object). You can use regression to fit a line to have the robot follow the wall/object.
I’m not sure of the ultimate use case, but I do see it being helpful for robotics.
The other might be weather/temperature prediction based on previous sensor readings in the past.
I am working on multiple room medical fridge. With the help of multivariable regression I want to monitor the frequency of how often the fridge is being opened and from there i will be able to predict the time period remaining for a particular room to go out of the acceptable temperature range. However, I do have three questions:
-How to upload a dataset with multiple labels?
-how to deal floating values for labels?
-how to identical labels?
Hi @josephsoso, your second and third question I answered in the other thread.
-How to upload a dataset with multiple labels?
We currently only support a single label per sample (for regression). If you want some multi-stage system you can create two projects, one with the first stage (f.e. raw data => predicting fridge open), and then a second stage (fridge open frequency => room out of temperature range), but I wonder if that’s actually needed. Couldn’t you label the data with ‘time period before the room goes out of acceptable temperature range in minutes’ (or something) and train on that? Then you pass raw data in and get that prediction out.
