20
Spent 4 hours troubleshooting a machine learning model because I forgot to normalize the input data
My accuracy was stuck at 45% and it turned out the raw pixel values between 0-255 were messing everything up, has anyone else wasted way too long on a simple data prep step?
3 comments
Log in to join the discussion
Log In3 Comments
kelly3851mo ago
Oh man, that's brutal... how did you finally catch it?
4
Been there with RGB values. Normalization gets skipped way more than people admit. Every time I learn this lesson again it hurts.
4
avery_flores171mo ago
Hang on, did you catch it by printing out a few sample values or by checking the loss curve? I'm trying to figure out if there's a faster sanity check than babysitting the preprocessing pipeline. Usually staring at a single misclassified image is what tips me off.
3