Hi Swadesh, Thank you for nice words and suggestion. Glad to know that this Matrix Factorization Recommendation system video helped you to learn. Keep Learning !!
at 9:23 , you say we have 15 elements but actually it's 12 which is smaller than the total of the factorized matrices (which is 14). How is this better? or am I missing something?
@@pathfinder9547 For much larger sets of data, for example 1000 users and 1000 items being rated, UserXItem ratings matrix would be of size 1000000. But if you do matrix factorization with ~30 latent features you could get 1000*30 + 1000*30 = 60000 size matrix which would greatly save space to represent essentially the same matrix. I think in the video because it is a toy example this logic was not clear.
@@pathfinder9547 im not sure what '70' you're talking about. But the 1000*30 matrix of users*features and 30*1000 matrix of features*items will matrix multiply to represent the full 1000*1000 matrix, so very little information is lost in matrix factorization via gradient descent. The 'factor' matrixes map the complex relationships between users and items via these 30 generated features. The '30' number can be increased or decreased depending on the complexity of the relationships. A more complex relationship will have more latent features, and simpler could have very few latent features. The number of latent features is a hyperparameter which can be optimized for finding the best sized factor matrices.
You are welcome Chhavi. Good to know, this Matrix Factorization Tutorial helped you. This is one of the my favourite videos, I had put a lots of effort to add many more concept in one video. Keep Learning !!
Thank you Rajan for nice words and suggestion. Will try to upload one video with Python code and really happy to know that this Matrix Factorization Recommendation system video helped you to learn. Keep Learning !!
Your smile while illustrating very complex idea is really really reliefing
Thank you for your simple and effective way of teaching . I'm a beginner but now i have a better view of recommandar system .
At 9:24 how did you get 15 elements? please elaborate
I think he assumed he had 5 columns instead of 4
Thank you so much sir, i got full clarity on matrix factorization. and your explanation is very easy to understand
Thank you for nice words. Keep learning !!
Good explanation. Thank you! Please do a full end-to-end recommender system model with python using this technique.
Hey Binod. Loved it. Keep up the great work.
Sir, You're Amazing!! Thanks for the explanation!
Brilliant explanation! Keep going ..
Are 15 elements kaha se aaya ?
you are awesome sir. one of the best explanation
Thank you so much! it was very helpful!
I really like the way you explained it... Good job. Thank you
wow, amazing explanation thank you so much, Sir.
Nice explanation sir👍
Glad to hear Vinnakota, this Matrix Factorization video helped you. Keep Learning and thank you for your nice words !!
super sir
Thank you sir. I love the way you explain
Great Explanation Sir !!
Hi Swadesh, Thank you for nice words and suggestion. Glad to know that this Matrix Factorization Recommendation system video helped you to learn. Keep Learning !!
at 9:23 , you say we have 15 elements but actually it's 12 which is smaller than the total of the factorized matrices (which is 14). How is this better? or am I missing something?
bhai answer mila iska? mujhe bhi yahi doubt hai.
@@pathfinder9547 For much larger sets of data, for example 1000 users and 1000 items being rated, UserXItem ratings matrix would be of size 1000000. But if you do matrix factorization with ~30 latent features you could get 1000*30 + 1000*30 = 60000 size matrix which would greatly save space to represent essentially the same matrix. I think in the video because it is a toy example this logic was not clear.
@@sakahamuru These 30 latent features should be linearly independent,right? The rest 70 would be a function of a few of these features ?
@@pathfinder9547 im not sure what '70' you're talking about. But the 1000*30 matrix of users*features and 30*1000 matrix of features*items will matrix multiply to represent the full 1000*1000 matrix, so very little information is lost in matrix factorization via gradient descent. The 'factor' matrixes map the complex relationships between users and items via these 30 generated features. The '30' number can be increased or decreased depending on the complexity of the relationships. A more complex relationship will have more latent features, and simpler could have very few latent features. The number of latent features is a hyperparameter which can be optimized for finding the best sized factor matrices.
In 11:23 , he Clear elemnet matter where right side have less element han left side.
thanks sir
You are welcome Chhavi. Good to know, this Matrix Factorization Tutorial helped you. This is one of the my favourite videos, I had put a lots of effort to add many more concept in one video. Keep Learning !!
where is next code video not getting
Nice explanation.
greate
Super sir..
thank you
Good explanation sir!
Happy to hear Abrar Amir, this Matrix Factorization videos Tutorial helped you. Keep Learning !! @binodsumanacademy
great job done!!!
do with python code
Thank you Rajan for nice words and suggestion. Will try to upload one video with Python code and really happy to know that this Matrix Factorization Recommendation system video helped you to learn. Keep Learning !!
tqvrm