You're the only person that explained this in a manner that allowed me to legitimately understand these topics. Rlly helping me out in my CIS class. Thanks a lot!
Lovely video!! I needed this for my exam. Can you please try to answer the questions you asked at the end? here are my guesses: Which model will be effected by missing data: Descriminative Which model will need more data: Descriminative. Less data: generative which model will be effected by outliers: i guess both? which model will need more calculus: I think Descriminative which model will tend to overfit: descriminative as well. Please feel free to answer and correct with simple explanations as soon as you can as my exam is approaching. I very much appreciate this! new subscriber:)
I think there should be some modifications: - Outliers have a greater impact on generative models due to the smaller amount of data points included. - Because generative models obtain the distribution of current data and examine it to the most likely distribution, they require more mathematics than discrete models.
Discrimivative models need more data therefore tend to be overfitted whereas Generative models built with less data may not generalize well with new data due to bias.
Outliers will affect only Discriminative right? or both? Both seems to be the right ans and my logic for it is that we already know that linear models see their curves affected by outliers while in the case of an algo like gaussian nb, the likelihood of an outlier happening will be very low for the given distribution and so that will bring the probability down. Can u please confirm if I am right or wrong?
please make a video on Generative adversarial network on regression problem. There are so many GAN models for Images, but i couldn't find one for continuous values
Hi Aman, I just started using Python. I am very basics. Please tell important functions that's very much needed for data scientist. Or tell where I can learn Python in advance level. Note: i am unemployed.
very good video , can you clarify below query the concept of generative mode is not clear in your example alien2 - compared features and did prediction alien1 - drew apple and banana and compared it with test sample and did prediction but to draw apple and banana we need to know its features correct then only you can draw it correctly so both models use features in the end to make prediction , so what difference is here ? how generative mode approach is different from discriminative ?
Generative models don't draw features infact they understand distributions. Whenever a new query point comes, based on the probability, the class with highest probability will be assigned like probability of a mail P(spam) = 0.4 & p(ham) = 0.6. The query point gets assigned to ham class.
@@billaspiel I think it's like this, In generative, we use features to find the distribution of the data in the n-dimensional plane. For example distribution 1 is for apple and distribution2 is for banana. When a new point will come we will measure the probability of this new point on those two distribution. and in discriminative, we use features to make the prediction directly. What is the best prediction for y given these x? Here we use decision boundary, not the distribution. For example, if the width is 10, the height is 5, color is yellow then it's a Banana.
You're the only person that explained this in a manner that allowed me to legitimately understand these topics. Rlly helping me out in my CIS class. Thanks a lot!
Thanks a lot.
Nice Explanation Sir
Thanks
Very helpful sir
You simply saved my life. Thanks!!!
Very clear explanation. I like the example and the visualization! I am a new subscriber!
Lovely video!! I needed this for my exam. Can you please try to answer the questions you asked at the end? here are my guesses:
Which model will be effected by missing data: Descriminative
Which model will need more data: Descriminative. Less data: generative
which model will be effected by outliers: i guess both?
which model will need more calculus: I think Descriminative
which model will tend to overfit: descriminative as well.
Please feel free to answer and correct with simple explanations as soon as you can as my exam is approaching. I very much appreciate this! new subscriber:)
Good answers Fatima.
I think there should be some modifications:
- Outliers have a greater impact on generative models due to the smaller amount of data points included.
- Because generative models obtain the distribution of current data and examine it to the most likely distribution, they require more mathematics than discrete models.
Explanation was so good! Also the quiz at the end, wow!
Nicely done
Thanks Rishabh.
Appreciate this explanation! TY!
i am regretting now for joining college wasting lakhs of money to learn nothing..but 5 mins u made the concept easy to understand...Hats off sir..
Great explanation! Simple and to the point. Thanks a lot! :)
Thank you.
Good explanation
Many thanks from Belgium!
Thanks for watching.
Thanks for making us understand in such an easy way ✨🙏
My pleasure 😊
Bhai i like your videos, I wish you grow on youtube .
Thanks Gotham.
Thank you! Awesome video, really great analogies and very clear.
Glad it was helpful!
Thank you for your clear explanation Aman👍
My pleasure
finished watching
great video
Thank you
Thank you Sir.. Nice explanation
Welcome Abhishek.
Discrimivative models need more data therefore tend to be overfitted whereas Generative models built with less data may not generalize well with new data due to bias.
Yes we can say like it Shobha.
Thank you
logit is definitely more prone to overfitting as it relies on more data to learn and there's a good probability that it will fit to noise
Outliers will affect only Discriminative right? or both? Both seems to be the right ans and my logic for it is that we already know that linear models see their curves affected by outliers while in the case of an algo like gaussian nb, the likelihood of an outlier happening will be very low for the given distribution and so that will bring the probability down. Can u please confirm if I am right or wrong?
please make a video on Generative adversarial network on regression problem. There are so many GAN models for Images, but i couldn't find one for continuous values
What kind of background you need to get into Data science or AI concept?
Statistics + Mathematics to start with
Hi Aman, I just started using Python. I am very basics. Please tell important functions that's very much needed for data scientist. Or tell where I can learn Python in advance level. Note: i am unemployed.
Just go thru the Code with Harry UA-cam channel.finish python.. rest will follow
The way of explanation is too good and the questions you asked, in the end, make me think deeply about what I understood.
Thank you
You are most welcome
Hi sir,
Can you take video on real-time A/B testing at the time of model deployment?
ok
Hi, can you make video for HMM model for Time series dataset?
So if my eyes are closed and someone gives me a piece of fruit and ask to taste it and tell me what it is. That would be discriminative?
Sir please deep generative model pr vedio bnaiye
very good video , can you clarify below query the concept of generative mode is not clear
in your example
alien2 - compared features and did prediction
alien1 - drew apple and banana and compared it with test sample and did prediction but to draw apple and banana we need to know its features correct then only you can draw it correctly
so both models use features in the end to make prediction ,
so what difference is here ? how generative mode approach is different from discriminative ?
Generative models don't draw features infact they understand distributions. Whenever a new query point comes, based on the probability, the class with highest probability will be assigned like probability of a mail P(spam) = 0.4 & p(ham) = 0.6. The query point gets assigned to ham class.
@@Julaiarvind. Thanks but to build a distribution we use frequency of a particular feature so wats the difference .
@@billaspiel I think it's like this,
In generative, we use features to find the distribution of the data in the n-dimensional plane. For example distribution 1 is for apple and distribution2 is for banana. When a new point will come we will measure the probability of this new point on those two distribution.
and in discriminative, we use features to make the prediction directly. What is the best prediction for y given these x? Here we use decision boundary, not the distribution. For example, if the width is 10, the height is 5, color is yellow then it's a Banana.
answer is Discriminative model will be effected by missing data.
True, thank you.
0:45 he asked u back What is Fruit. 😂😂😂😂
🤣🤣
great video
Thanks Aditya.