Need help with my UFC fight prediction Tabular Model

Hey everyone!

Going through the corse for the first time and wanted to play around and create a model to predict ufc fights for fun.

Here’s a link to my colab

The tab I’'m trying to predict is the column “result” and every row of the csv represents nearly every fight in ufc history and theres data for each fighter.

My current issue I’m getting is that for each epoch i train it says accuracy is 1.0 which is impossible so I’m wondering how I could debug this. I assume theres some data leakage or maybe I’m using too many columns if thats an issue?

Would appreciate any help at all even just some paths I can read up on so I can figure out some progress



The Colab notebook achieves an accuracy of ~50%, perhaps that is what you had in mind by saying the model reaches 100% accuracy? In either case, in the code, fastai mistakenly assumes the task at hand is regression, thereby incorrectly setting the net’s output size to 1 and the loss to mean squared error. To specify that this is a classification problem, please pass y_block=CategoryBlock() to the TabularPandas object; that ought to fix the issue.

1 Like

Thanks so much! What a simple thing to miss,

Really appreciate it. Still have a long way to go but no longer stuck

1 Like