Also in Principal Component Analysis, scaled features are very important because we search for the principal axes that have the highest variance. So if we have one feature in [0,1] and the other one in [1, 100], then the latter one has a much higher variance, even though it may not contain much information to be kept by the PCA.
How many more people would understand math if we had explanations like this. I feel like I have been reading math papers written in French, and you just spoke in English for me. Gosh, THANK-YOU.
I was wondering where you’ve been! Nice to see you back to posting. Well covered topic - it’s easy to overlook standardization and normalization thinking they are simple. They have some important subtleties
Great, specially good to explain the misconception with non linear transformations which for some reasons is constantly used in conversations as normalization/standarization
So much of machine learning is not novel. It is merely "rediscovered" classical statistics approaches with a new marketing language. The example he shows at the end is essentially principal components using correlation instead of covariance because 2 metrics have vastly different variances or scales.
Good video - your description and explanation is good. However relating the basic explanations to real world problems would be helpful for users. Also using a partial distribution to calculate things such as volatility based on only the negative change is interesting. Also using curve fitting of data to determine parameters for trading and models is also interesting
Very good. I have a doubt. I would love to hear your comment on it. In recent months, I have been reflecting on the apparent prevalence of certain predatory mega-journals, in particular MDPI's Sustainability, which stands out as the journal with the most publications on various topics, according to various tourism bibliometrics. However, this observation has led me to consider the need for further analysis. Specifically, it has caught my attention that when using the percentage of publications in relation to the specific research topic in percentage terms (number of articles on a topic divided by the total number of articles published), the magnitude of the contribution decreases drastically. To illustrate this point, let me present a hypothetical example: Journal A has published 10 articles on prospect theory in the last five years, but its total output is 600 articles. In comparison, Journal B has published 25 articles on prospect theory in the same period, but its total publication volume exceeds 49,000 articles. Some bibliometrics would say that Journal B is the one that publishes the most, however, it is just a matter of gaining by quantity. I gave the journals weights based on their percentages (Weight of journal = Percentage of Journal / Highest Percentage among journals) then I did the min-max normalisation (Normalised weight = (Weight of Journal−Min Weight) / (Max Weight−Min Weight)), Then I created a Weighted Metric with Normalisation (multiplying the normalised * their weight). The use of min-max normalisation in this one is correct? Do you think there is a better approach?
Jesus, thats so great. Im totally new to data science and ML and Im trying to take it slow to properly understand everything. This video was super great in doing that. I picked up new knowledge that will be helpful for when Im writing my own ML algorithm (probably KNN based image classfication)
Hi there thanks a lot! I have one question on min-max normalization as I m using Stata. When I use the formula, shall I take into consideration the actual min and max values of the variable, or I should consider the potential/feasible range of values the variable can assume? E.g. I have one variable that can take values -100,+100, yet in my dataset the min is -12 and the max is 34.
Thanks man for the video. this was with no doubt very helpful. however i was wondering how do you make all these animations ? Thanks in advance for you kindness.
Hi. For deep learning, it best to do min-max normalization (i.e. stretch values to 0-1) or max normalization (i.e. only divide by max to keep within 0-1)? I see a problem with the former approach, as a single outlying value can significantly skew all the rest of the values, making them not very comparable to the reference values.
i'm new to machine learning and theres something i dont quite understand: if you scale the X(input), does it affect the Y(output)? In a real life scenario where i want to make a prediction with my model, wont the scalling affect the results? if i shrink the input wont the output also be smaller?
By looking at what you are saying: No, I don't think so (don't take my word though, I am new at ML). I'd say your weights will be computed accordingly. But I read that even scaling your outputs (before the training) is a thing, there are people who do that.
Can someone here help me with my data preprocessing project or know where i can find help? I am so stuck and cant get over 70%. i really wann do well but dont really know what else do in preprocessing
Bro, but how can we decide which technique to use when? and if selecting normalization then which normalization such as----min-max etc.....? could you please elaborate this.
Ahhhahaaa, I was sad seeing your last video was a year ago. Your visualization is really cool and as good as intuitive ml. But he stopped making videos 3 years ago
This video should be nominated to the UA-cam Oscars/Grammy awards....
Also in Principal Component Analysis, scaled features are very important because we search for the principal axes that have the highest variance. So if we have one feature in [0,1] and the other one in [1, 100], then the latter one has a much higher variance, even though it may not contain much information to be kept by the PCA.
Great point! Feature scaling is very important in pca also.
How many more people would understand math if we had explanations like this. I feel like I have been reading math papers written in French, and you just spoke in English for me. Gosh, THANK-YOU.
Your clarity is amazing. This helps! Sub earned
This is the first video I watched and man you have crushed it. This intuitive explanation of math was a joy to watch. Please keep them coming.
Beautiful explanation and editing. Well done
I was wondering where you’ve been! Nice to see you back to posting.
Well covered topic - it’s easy to overlook standardization and normalization thinking they are simple. They have some important subtleties
I saw you today in Yannic's channel as well, nice to see you again.
Thanks a lot mate! Really happy to be able to upload again :D❤️
Hey DJ, we are waiting for you also!
@@taotaotan5671 lol coming soon!!
A standardization makes the original distribution look more normal . It doesn't just make a zero mean and 1 stdev.
Man this is the most intuitively explained video for this topic i ever found ,thanks man
Great, specially good to explain the misconception with non linear transformations which for some reasons is constantly used in conversations as normalization/standarization
You're doing amazing work here, hopefully one day you will get the recognition you deserve
Great videos, dude!
It's a shame we no longer get this great content
:(
Normalisation became new normal to me, great job dude!!!!
We always learn something new even when I knew the topic from before! you make it more interesting and valuable. Thanks from Kuwait
Very nice explanation and demonstration. Good topic.
So much of machine learning is not novel. It is merely "rediscovered" classical statistics approaches with a new marketing language. The example he shows at the end is essentially principal components using correlation instead of covariance because 2 metrics have vastly different variances or scales.
thanks man, It's help me so much to understand about normalization
Very helpful
Thanks to you I understood why feature scaling is imp, thank legend
I just love your channel name so much
sir olease make more videos, your sessions are very helpful
Very nice video! Everything became clear as soon as I watched this
Saludos desde Perú, Excelente tus videos! gracias! éxitos.
Great explanation boss helped a lot chaliye jaao guru
Superb explanation
Really nice explanation
Good video - your description and explanation is good. However relating the basic explanations to real world problems would be helpful for users. Also using a partial distribution to calculate things such as volatility based on only the negative change is interesting. Also using curve fitting of data to determine parameters for trading and models is also interesting
This guy explained something my lectures failed in years, in 5 minutes
such an informative video! good work!!
Great to see you back bro ! ✌️
Thanks a lot!! :D
Wow - what an incredibly helpful video; thank you.
wow bro, this was really really good. Thank you soo much.
Very good. I have a doubt. I would love to hear your comment on it.
In recent months, I have been reflecting on the apparent prevalence of certain predatory mega-journals, in particular MDPI's Sustainability, which stands out as the journal with the most publications on various topics, according to various tourism bibliometrics. However, this observation has led me to consider the need for further analysis.
Specifically, it has caught my attention that when using the percentage of publications in relation to the specific research topic in percentage terms (number of articles on a topic divided by the total number of articles published), the magnitude of the contribution decreases drastically. To illustrate this point, let me present a hypothetical example:
Journal A has published 10 articles on prospect theory in the last five years, but its total output is 600 articles.
In comparison, Journal B has published 25 articles on prospect theory in the same period, but its total publication volume exceeds 49,000 articles.
Some bibliometrics would say that Journal B is the one that publishes the most, however, it is just a matter of gaining by quantity. I gave the journals weights based on their percentages (Weight of journal = Percentage of Journal / Highest Percentage among journals) then I did the min-max normalisation (Normalised weight = (Weight of Journal−Min Weight) / (Max Weight−Min Weight)), Then I created a Weighted Metric with Normalisation (multiplying the normalised * their weight). The use of min-max normalisation in this one is correct? Do you think there is a better approach?
So glad to see you back !
So happy to hear that :)
thanks man. amazing video
Great lesson! Thank you so much for you video
thank you so much for this!
Love it when you have on your glasses
great video, to the point with great visuals, subscribed.. Btw, how did you make these nice graphics?
can you pls respond ?
WOWW! Absolutely loved this! Thanks
Jesus, thats so great. Im totally new to data science and ML and Im trying to take it slow to properly understand everything. This video was super great in doing that. I picked up new knowledge that will be helpful for when Im writing my own ML algorithm (probably KNN based image classfication)
High quality content. Thank you!
Good Explaintion... thank you very much 😊😊😊😊😊😊
I LOVE YOUR VIDEOS
Very good explanation.
Absolutely loved the explanation!
So glad!
How can you so perfect in explaining
coolest presentation!
can you please make a video on PQN normalization?
Great to get back nerdy notifications...
:D :D
Great video!
Good that you are back!😎
Thanks!! 😁
May I ask about the technologies that have been used to create this content ?
I really appreciate sharing.
good explanation
An excellent explanation...Thanks a lot for sharing ....
Excellent! Thanks.
Thank you!
Excellent visuals!
Love the sound effects! lol
extremely beautiful viz , teaching methodology is amazing too. I too run ana analytics channel, but u inspired me more
AWESOME VIDEO TYSM YOU'RE AWESOME
thank you so much
excellent visualization, thanks!
love it. thanks so much for the explanation
Hi there thanks a lot! I have one question on min-max normalization as I m using Stata. When I use the formula, shall I take into consideration the actual min and max values of the variable, or I should consider the potential/feasible range of values the variable can assume? E.g. I have one variable that can take values -100,+100, yet in my dataset the min is -12 and the max is 34.
Thanks man for the video. this was with no doubt very helpful.
however i was wondering how do you make all these animations ?
Thanks in advance for you kindness.
Your explanation was damn neat!
great video
thank you
Hey! I wanted to know which software/ tools you used to make videos like this?
Good video, content animation are amazing.
Very helpful
Hello brother. If i have given a time scale for a Analysis. Can i use this time scale to normalize my analysis time?
Yayyyyy! Thanks for an amazing video.
😁😃
I really hope you are fine now. Your videos helped me a lot in several times. Easily you could be a teacher if you want to. Thanks!
could you please tell me what software you used for these visualizations
Great videos! May I ask what software you use to create your equations/animations?
I think he uses manim
Hi. For deep learning, it best to do min-max normalization (i.e. stretch values to 0-1) or max normalization (i.e. only divide by max to keep within 0-1)? I see a problem with the former approach, as a single outlying value can significantly skew all the rest of the values, making them not very comparable to the reference values.
superb !
do you use manim?
Please add NLP course.
Hey, have you checked this playlist?
ua-cam.com/play/PLM8wYQRetTxCCURc1zaoxo9pTsoov3ipY.html
Feel free to suggest more topics!
yes ty
You are the best!
what software do you use for animations?
thanks bro
Omggggg ur back!!!
Yeaaah ❤️🥹
Well explained.
Thanks man!
Very nice!
i'm new to machine learning and theres something i dont quite understand:
if you scale the X(input), does it affect the Y(output)? In a real life scenario where i want to make a prediction with my model, wont the scalling affect the results? if i shrink the input wont the output also be smaller?
By looking at what you are saying: No, I don't think so (don't take my word though, I am new at ML). I'd say your weights will be computed accordingly. But I read that even scaling your outputs (before the training) is a thing, there are people who do that.
excellent.
Amazing explanation! Thank you.
The datasets get normalized just like the speaker! (a joke, couldn't help it)
Excellent!
Many thanks!
Amazing amazing amazing!!!!!
Can someone here help me with my data preprocessing project or know where i can find help? I am so stuck and cant get over 70%. i really wann do well but dont really know what else do in preprocessing
gg budd you opened new horizons for me
Bro, but how can we decide which technique to use when? and if selecting normalization then which normalization such as----min-max etc.....? could you please elaborate this.
♥️♥️♥️
❤️😍
Ahhhahaaa, I was sad seeing your last video was a year ago. Your visualization is really cool and as good as intuitive ml. But he stopped making videos 3 years ago
Sorry, I can't understand at 3:10 : Good old [what?] algorithm
Gradient Descent Algorithm
Now I realize I'm really poor in math.😥
Noor and malai become normalization 😂😂😂since noor and malaise are top by nature 👌 😂😂
blud comes after 1 year and does not come back even after another year gone past .