This is a demonstration of how to create a model using BigQuery ML. After testing the model, we look at the results and discuss some of the reasons the predictions aren't more accurate.
When you removed those that are above 100 but didn't limit those below 0, the skew of the distribution made your predicted values to move tho the left, resulting in negative values and most of the predictions being below what expected. Nice video!
Hi Professor, great video as always! Do you think you could reexplain what you did at 4:00? It seems that you trained the model on data from before 2022 and now you are predicting the average time for 2022? In the video you say "train" again when using "ML.PREDICT" for the values in 2022 so I was slightly confused.
I think you've got it, but I might have misspoke on using "training" again. Basically, we are training the model on data before 2022, and then testing it on data in 2022. In a demo, it avoids the situation where the model already knows the answer; if we had trained it on 2022 data and then asked the model questions that it'd already been given the answers, then it wouldn't be a super useful demo. Typically we separate data into training and validation sets - it's kind of like that.
