Why Does ML.NET Value Prediction with R2 0.7000 Always Give the Same Value Prediction?

30 Views Asked by micahhoover At 19 February 2024 at 19:36

I'm only interested in the title and the views (I'm trying to see which titles work better than others).

I wrote some custom C# to deal with duplicative commas in the CSV. You can eyeball the screenshot to see it at least seems to work OK.

Then I trained it a ML.NET Value Prediction model for 600 seconds and it completed in less than 500 seconds.

In the evaluate tab, it uses LbfgsPoissonRegressionRegression as the Model and has an R2 value of 0.7000. Which looks ... good ?

But then when I try to get predictions, every title has the exact same prediction : 196,163.11 views.

I'm not sure why this is far less reliable than the YELP example I tried earlier with the sentiment analysis.

What could possibly be breaking this ? Should I use an alternative to the "Value Prediction" in the Add Model wizard in Visual Studio ?

There are 0 best solutions below