Quick question, it’s obvious that normalizing training data set will improve neural net performance overall; but do we have to normalize our input as well when we are predicting using the trained model?

Yes, you do. Your model is expecting that kind of input modification

Thankyou, you’re the best!

Btw, how do you do this with just one row of data?

Currently I’m using the Normalize class for continuous data and Categorify for categorical data.

If using v1 you need to create a TabularList (IIRC) and pass this to load learner. In v2 you use dls.test_dl(your dataframe). See the inference docs, there’s a tabular example. (

Is this how?

holdout is a testing set separate from training and validation set