Great video, thank you so much! I don't understand one part though: will nDCG be the same for when we guessed 2 items out of 10 guesses and have them in the perfect order versus we guessed 5 items and have them in the perfect order? Shouldn't the "perfect order" include guessing more correct items? Meaning that if there are 5 possible relevant guesses, and we only guessed 2, IDCG calculation would have all 5 in the perfect order vs. having only two in the perfect order?
In short for this video: use NDCG score to evaluate your recommendation model. I suppose the model is a ranking model (Learning-to-Rank), it is not a traditional classification or regression model.
It was helpful to understand DCG and NDCG - but I would also add what you sad at 9:04 in the beginning - hard to replace off line as such with online and this only help to prioritize what to test the next.
Can't even get your prices right, find duplicate listings, use nonsensical naming conventions and what's up with your SKUs? And ten thousand dollar throw pillows, wtf. Also, you recommended me the exact same sofa I was looking at, but at a different price. Some data science, smh.
The best NDCG explaining video I've seen so far! Thanks!!
thank you million times! after reading tens of articles I finally understood DCG watching this video. super clear
Great content! Short and sweet.
Great video, thank you so much!
I don't understand one part though: will nDCG be the same for when we guessed 2 items out of 10 guesses and have them in the perfect order versus we guessed 5 items and have them in the perfect order? Shouldn't the "perfect order" include guessing more correct items? Meaning that if there are 5 possible relevant guesses, and we only guessed 2, IDCG calculation would have all 5 in the perfect order vs. having only two in the perfect order?
can we please have more videos on recommendation systems ?
In short for this video:
use NDCG score to evaluate your recommendation model. I suppose the model is a ranking model (Learning-to-Rank), it is not a traditional classification or regression model.
It was helpful to understand DCG and NDCG - but I would also add what you sad at 9:04 in the beginning - hard to replace off line as such with online and this only help to prioritize what to test the next.
How do you evaluate seriously overpriced cabinets named after missing kids?
Can't even get your prices right, find duplicate listings, use nonsensical naming conventions and what's up with your SKUs? And ten thousand dollar throw pillows, wtf. Also, you recommended me the exact same sofa I was looking at, but at a different price. Some data science, smh.