nDCG: the evaluation metric you've (probably) never heard of

[@AnkitSharmaKumar]

Brilliant stuff. Have been following your channel for quite some time and I must say your videos are among the best content on ML. Keep up the good work 👏

[@ritvikmath]

Thanks :)

[@LeonLokk]

Very interesting! I never knew there was such a metric for ranking things. Glad you explained the intuition behind the formula - helps so much in understanding what is going on :)

[@spicytuna08]

thank you very much. I looked every where for this. there are so many places explain about ranking machine learning, but never explained about how to evaluate. some project did rmse evaluation on cos similarity machine learning. i could never understood that. this confirms that rmse evaluation is not possible for ranking machine learning. thanks again. well done.

[@PrabhuRamNagarajanES]

How to obtain relevance doc without cosine similarity? please explain

[@qiguosun129]

Always great talks, thanks for sharing!

[@DYMLikeABaws]

I loved this explanation and the fact that you explained the formula

[@WatermelonSaurus]

wow. this is what i've been curious about all the time. thanks for the video.

[@hangchen]

So good explanation. Watched once and understood nDCG.

[@ncarrillox]

Thank you for this explanation. Math isn’t my strong suit and you made this video very approachable. Subscribed.

[@toquyenphun]

Thank you so much for your explanation! Keep up good work!

[@NitinMuteja-ki9oq]

very helpful. Thanks for explaining this with simplicity.

[@Set_Get]

Good as always. Thank you

[@thibaultguittet577]

Interesting and easy to grasp video as always. I prefer this format to the paper sheets !

[@xiaoruizhou694]

great talks, thanks for sharing!

[@anonymousop7512]

Amazing explanation!

[@PrabhuRamNagarajanES]

Good Explanation sir... Could you tell me how cosine similarity difference from Ideal Discounted Cumulative Gain. How the order of relevances are decided? whether it is randomly sorted or by other mean. Please explain more...

[@pierreroux-lafargue8227]

Hello, have you used sklearn.metric.ndcg_score by any chance before? With your simple example sklearn returns different ndcg scores if we simply pass in your two arrays np.array([2, 1, 0]) and np.array([0, 2, 1]) as an example. Do you know what input needs to be fed ?

[@fengjeremy7878]

Very very clearly explainded!!!

[@keyuchen3738]

great job, very clear

[@sriks4003]

Thank you, very helpful!

[@moredeathstar]

How to calculate the relevance score of a document/item that I am recommending?

[@rodicacoderie4049]

Great content, following your channel for some while now and it helped me a lot! You have a great way of explaining hard concepts. Can you please make a video about ways to model and evaluate zero-inflated distributions? I know that Tweedie regression is a possibility, but the explanations I found are not that intuitive. Thanks and keep going!

[@gpprudhvi]

I think I'll get a reply for this comment. Hi Rithvik, thanks for everything you are doing. Your explanations are as clear as the word transparent.

[@ritvikmath]

Thanks :)

[@jayyee3790]

This was really helpful! I'm building out a similar model at work and just starting to wrap my head around the concept of LTR. One of the things we're trying to implement is making our recommendations more diverse. Have you ran into any practical concepts of this before (e.g. trade result accuracy for diversity)?

[@KolliVenkataMadhukar]

so why dont we just consider the values instead like 210 > 102 > 021 why we need to define this metric in the first place, might be a silly doubt but unable to figure out the exact reason

[@arsnakehert]

Thanks for the explanation However, this is a measurement for the ranking, right? And it seems to count on the documents already being classified as relevant or not? I'm wondering, how is the relevance of an individual document for a search query measured? One thing that went through my mind is aggregating, let's say, a database table with columns such as the search query, a document ID, and how many times it was clicked from that query, or something like that. Would this be a reasonable relevance measure for a single document, which could then perhaps be used for scoring search result relevance?

[@AlshehriAbeer]

amazing! easy to follow up with him

[@pgbpro20]

Move over Andrew NG, Rithvik has arrived!

[@ritvikmath]

Lol 😂

[@AlshehriAbeer]

I'd recommend change the name of the vid or add to it the main object of the vid, I went crazy to find this video again

[@ArunKumar-bp5lo]

great Video but i can't find which playlist this video belongs to ??

[@ResilientFighter]

A ritvik video a day, makes a better data scientist for another day. (Not good with rhymes okay) but good job ritvik!

[@chd9841]

Where is video for listwise ranking?

[@ritvikmath]

On my list 🫡

[@ziv1153]

How do you determine the log integers? Is it always log2(N) or it's the log of the max ranking (Nmax)?

[@sorrefly]

That would be cool if you could make a video about the gaussian process using kernels :)

[@sa-vx5vi]

Thanks!!

[@bhoomeendra]

Thanks for the explaning but I belive that NDCG = DCG/IDCG is easy to understand

[@akshathashashank801]

Very well explained