While I find the CS50 video player on their website to be somewhat inconvenient due to the lack of hotkeys, I appreciate the ease of watching the lectures on RUclips, still utilizing the CS50 website for quizzes and projects.
pomegranate v1.0.0 is a ground-up rewrite of pomegranate using PyTorch as the computational backend instead of Cython. API is significantly different. unable to understand use v1.0.0.. as lot changed
1:23:47 if you're just gonna calculate probability like this (6/8 in this case) without using any of the probability values then why even sample? just calculate all possible combinations of the other 3 variables divided by possible combinations of all 4 variables using permutations and combinations.
I cannot seem to import pomegranate correctly. The library seems to be installed but having NameError for all classes used in the source code. Spent two days just to figure out how to make it work. Does anyone having a similar problem? And would using an earlier version of Python be a likely solution?
about the Bayesian network slide, point 3: arrow from X to Y means X is a parent of Y, point 4: each node X has probability distribution P(X|Parents(X)), should it be written as P(Y|Parents(X)) instead?
Those two X's aren't related, the X at point 4 is any node that has a child Y. While the X at point 4 is any node that has a parent. So for the example in point 3 where X is the parent of Y, the probability of Y is written as P(Y | Parents(Y)) where Parents(Y) is X.
I also have the same question. I think it is conditional probability because the marginal probability is not 1. If you add all rows and columns up, it should be 1 for a joint probability distribution table.
I don't understand the reason for sampling, when you have the probabilities, why wouldn't you just calculate the solution instead of drawing samples based on the probabilities? Doesn't that need more computation power and give worse results?
Will it need more computational power/time though? I guess we randomly have to assign any of the possible values that a particular random variable can have and do that for all the random variables, to get a sample. The time required for creating one such sample is linear and then all we have to do is search each sample for the required queries which in itself is a linear operation.
Because this is just a simple example. In some cases you are not able to "calculate" the solution, and the closest we have is estimating through simulation/sampling.
@@MrJoeDone you might not have them or could be the interaction of multiple variables with complex probability distributions that sometimes is not possible to calculate
This is just out of this world lecture. This stuff is always difficult in other courses/books, but Brian makes it so logic and clear. I bow. Thanks
The guy can barely speak, I fail to see how this is 'world class'.
Damn Brian just goes up there and bangs out a 2hr lecture like it's nothing
fr (for real)
I'm very curious about what his IQ is.
Cs50 just has the best
how do you comment 2 years ago if this video is released 9 months ago 💀
😂@@turtles8229
While I find the CS50 video player on their website to be somewhat inconvenient due to the lack of hotkeys, I appreciate the ease of watching the lectures on RUclips, still utilizing the CS50 website for quizzes and projects.
I was overwhelmed by Stanford AI classes and I am sooooo glad I found Harvard AI intro classes, you did best job Brian!
his teaching and presentation skills are excellent i wonder how i become like him
Such an amazing course! Thank you so much
Brian is indeed a great teacher.
at 26:10 I think the not-equals sign ought to have changed to equals once we are considering the conditional equation right?
Yeah
I think it’s a typo
one of the best lectures ,I have ever seen
Just the quality of this Lecture is mind-blowing. When I remember my time at the university in Germany, its just sad.
Germany has no good university, USA rule! Sorry you studied Germany🤥 Go to USA
I love this 💜
This is definitely one of the hardest lectures from CS50 to understand
This is definitely one of those tracks you listen to without falling asleep.
literally high school math. Thanks Brian for making it so simple!
26:15 That unequal sign should be equal sign :)
no
26:23
Joint probabilities on 35:45, is that correct? isn't P(cloud and rain)=0.4*0.1=0.04? enlighten me
P(cloud, rain) + P(cloud, !rain) = P(cloud)
0.08 + 0.32 = 0.4
(This 0.08 and 0.32 values is not calculated)
So proud spotted an error from this brilliant team at CS50 around 26:12 the not equal sign should go once it becomes dependent
dont have words to praise
pomegranate v1.0.0 is a ground-up rewrite of pomegranate using PyTorch as the computational backend instead of Cython. API is significantly different.
unable to understand use v1.0.0.. as lot changed
same here
how didiyou resolved it? facing the same problem?
lol didnt expect this to be here it was worst in my college 1st sem .. but this guy ..!!!
18:28 Random variables must be mutually exclusive and sum to 1.
P(sun) + P(sun & wind) + ... + P(cloud) + P(cloud & rain) + P(cloud & rain & wind) + ... = 1
1:23:47 if you're just gonna calculate probability like this (6/8 in this case) without using any of the probability values then why even sample? just calculate all possible combinations of the other 3 variables divided by possible combinations of all 4 variables using permutations and combinations.
thanks for amazing course
is there other sources where i can better understand all these, new to me, statistical concepts????
how he calculated the joint probability for the table of cloud and raining, given the dissociated probabilities? It is show in the time 35:15
I don’t think he calculated it
It was just an example of a joint probability
That is additional information. He didn't calculate it
I cannot seem to import pomegranate correctly. The library seems to be installed but having NameError for all classes used in the source code. Spent two days just to figure out how to make it work. Does anyone having a similar problem? And would using an earlier version of Python be a likely solution?
Same here, did you figure it out?
the joint probability stuff is weird and explained vaguely for me.
about the Bayesian network slide, point 3: arrow from X to Y means X is a parent of Y, point 4: each node X has probability distribution P(X|Parents(X)), should it be written as P(Y|Parents(X)) instead?
Those two X's aren't related, the X at point 4 is any node that has a child Y. While the X at point 4 is any node that has a parent. So for the example in point 3 where X is the parent of Y, the probability of Y is written as P(Y | Parents(Y)) where Parents(Y) is X.
55:25 shouldn't that be joint probability instead of conditional probability?
I also have the same question. I think it is conditional probability because the marginal probability is not 1. If you add all rows and columns up, it should be 1 for a joint probability distribution table.
Bro Brian going in with this series
1:24:00 Isn't there an error? All the "attend" should be highlighted in green?
No, he is looking for the probability of the train being on time as to 1:23:30.
Shouldnt that be 24 for all options available (3x2x2x2) instead of 8 (as shown)?
Class and DiscreteDistribution seem to have been deprecated in pomegranate v1.0.0
same with me, is there any tips on how you resolved it?
How can we calculate the joint probability he showed at 36:12
It is given, he didn't calculate it.
Very cool
love you ❤❤❤❤❤
I don't understand the reason for sampling, when you have the probabilities, why wouldn't you just calculate the solution instead of drawing samples based on the probabilities?
Doesn't that need more computation power and give worse results?
Will it need more computational power/time though? I guess we randomly have to assign any of the possible values that a particular random variable can have and do that for all the random variables, to get a sample. The time required for creating one such sample is linear and then all we have to do is search each sample for the required queries which in itself is a linear operation.
Because this is just a simple example. In some cases you are not able to "calculate" the solution, and the closest we have is estimating through simulation/sampling.
@@wmaps but you still have all the probabilities even if it is a more complex cenario
@@MrJoeDone you might not have them or could be the interaction of multiple variables with complex probability distributions that sometimes is not possible to calculate
Hey need explanation about this, shouldnt that be 24 options available (3x2x2x2) instead of 8 options as shown?
What I've learnt from this is that Brian needs to get an earlier train to make his meetings on time 😂. Great lecture though.
I always feared when you go to university you wil be overflooded with data.
This kind of videos proves that to be wrong...
Bye bye fear....
😃
its just theory or can we see some practical application :(
Probability
Set theory
I have no idea what Brian is talking about ....
Same 😔
Thank you
playlist is about 2023 but in each title is 2020 why? is it typo?
❤
its nothing in stadistic
anyway its still use in the real life for use programming in your life old class
44:10
Fair dice, lol.
What an awesome lecture. Thanks 🙏🏽