In the very crowded field of RUclips videos and, literally, centuries worth of content out there now that could be watched, I still say you have the absolute best channel. The length of your videos are always just right - not too long, not too short - and with the exact right mix of theory, technical math/stats, visuals, examples, and breakdown on why something is useful and applicable is why it's my favorite channel. 🤓
You have a great ability to simplify complex topics into layman's launguage. Absolutely loved it. Would hope you talk a bit more frequently on the philosophical side of math. Kudos!!
Love it! Can bench watch your videos all day! Love how you break things down in layman perspective. It makes learning statistics so simple and relatable. Wish I knew about your channel earlier!
Dear, I really enjoy your videos, I'm easily carried away by your reasoning and I honestly appreciate your sense of humor. You are one of a kind. Please keep up!
Excellent articulation of your thoughts! Thank you. At 9:00, and in previous videos of yours, you mention the heart of Bayes. We all tend to think about and process our world with at least a touch of Bayes (I'd bet even die-hard frequentists do in a general setting too, outside of work/academics). The P(robbed) example with the two people moving to the new city is a perfect example; their initial behavior is quite different, informed totally by their priors. However, if they live in the city long enough (see enough data), they should start to converge on similar "what should I do each day to not get robbed" behavior I'm seriously enjoying your content, btw. I'm using Bayesian statistics in a research project of mine for grad school and your content is immensely helpful
Hey man love your channel and this video provided a lot of clarification for me. I think this discussion is a bit more applied than you imply. In my day to day data analysis, I'll be using the frequentist approach for the vast majority of the time, but now you have nudged my mind to consider whether a Bayesian approach might be useful in certain circumstances with limited data that makes the frequentist approach inapplicable. Basically, I can see applying a Bayesian approach in the right context being a great way to think laterally, get creative, and even generate interesting hypotheses that can possibly be tested later using a frequentist approach! This is tied to another concept you mention in this video; that of the Bayesian converging with the frequentist as data become available.
I’ve been looking all over for an explanation of the difference between frequentist and Bayesian statistics. Your video finally cleared it up for me, thank you so much for the quality explanations! Your channel has been feeding my growing interest in statistics 😄
Superb! I'm a "measurement uncertainty geek". You wrapped a ton of concepts into your speed of light example. Even if everyone concedes that Einstein was correct and that there is only one possible value for this constant, all attempts to measure it will produce a variety of results. Also, I belong in the camp that believes that it is a waste of time to even discuss the "true" value for this parameter because it just adds confusion and cannot ever be attained down here on Earth. All that we can reasonably expect is to narrow the range of possible values down to a point at which the possible variations is too small to bite us on the ass when we need to depend on one value for some other purpose. Thank for wandering a little distance from your normal, very practical starting point, which is so very powerful and trustworthy.
A common misconception is that Bayesian priors are subjective probabilites based on personal experience and so computed results are at the whims of individual beliefs. Priors in science are almost always objective probabilities drawn from current scientific knowledge to which observed data is then added via the likelihood to generate a posterior probability. Priors are no more subjective than choosing the method for acquiring frequentist data or choosing a frequentist model.
This was quite enlightening. I have to watch and read more about it though, because my gut impression, even after the video, is that Bayesianism just feels like bullshitting prejudice when you don't have the evidence at hand.
In some instances you cannot collect data or do the experiments. Examples from the cold wars were bombers carrying nukes and nuclear subs today. Another is search patterns looking for planes or ships lost at sea.
I really enjoy your work. You make it look all very easy, as it is really not. I am curious what your thoughts are about epistemic and aleatory uncertainties
Great video! I'm probably totally off on this, but it seems like frequentism benefits from thinking about a multiverse, because you can imagine that events which happen once in our universe actually have an objective long-run frequency when you consider them across all universes. So when asked to give a range of possibilities for the speed of light, they might consider all copies of themselves across other universes, some of which will live in universes with a different speed of light. It seems like it might be more of a practical limitation (can't do much with a sample size of 1 even if many other samples might actually exist) than theoretical.
Based off my previous experience with this channel I have a 90% like rate....so I'm going with that....don't need to sample all of your videos and then all the other videos that would have similar content and then feel confident to know that a thumbs up is warranted. But if the speed of light didn't have a normal distribution then....
thanks for this important topic, i never been satisfied with statistics and propbablity either their definitions or mathematical formulation, for example what the term "density" really means ?
That's not quite true. Both Bayesians and Frequentists have a notion of "a priori" probability of *events* . But they define it on different things. For Bayesians the data are fixed numbers while the model parameters are assigned probability distributions, while for the Frequentists the data themselves are samples from a population (hence sampling probability) while the parameters are fixed. *This* is the crucial difference. By treating the parameters as random variables, Bayesians can build hierarchical models, where the prior "hyperparameters" get hyperpriors (i.e. PDFs) as well. There are also various schools of thought *within* the Bayesian camp with respect to what probability is. Interpretations like de Finetti etc, having to do with exchangeability, symmetries etc.
In the very crowded field of RUclips videos and, literally, centuries worth of content out there now that could be watched, I still say you have the absolute best channel. The length of your videos are always just right - not too long, not too short - and with the exact right mix of theory, technical math/stats, visuals, examples, and breakdown on why something is useful and applicable is why it's my favorite channel. 🤓
Wow! That really means a lot. Thank you.
@@ritvikmath 100% agreed. Your mix of intuitive explanations and math is fantastic! It's so common to see explanations that ignore one or the other.
Just beautiful
Thanks for this
Thanks!
Thank you, this video really helps to look at probabilities in a way that makes it more understandable.
What a great video. Thank you so much for all the excellent content you create :)
Been following you for a long time. Looking forward to an #AMA (Ask Me Anything) episode.
You have a great ability to simplify complex topics into layman's launguage. Absolutely loved it. Would hope you talk a bit more frequently on the philosophical side of math. Kudos!!
Good suggestion! Thanks!!
Love it! Can bench watch your videos all day! Love how you break things down in layman perspective. It makes learning statistics so simple and relatable. Wish I knew about your channel earlier!
Glad you like them!
Dear, I really enjoy your videos, I'm easily carried away by your reasoning and I honestly appreciate your sense of humor. You are one of a kind. Please keep up!
The best channel ever!!
Excellent articulation of your thoughts! Thank you. At 9:00, and in previous videos of yours, you mention the heart of Bayes. We all tend to think about and process our world with at least a touch of Bayes (I'd bet even die-hard frequentists do in a general setting too, outside of work/academics). The P(robbed) example with the two people moving to the new city is a perfect example; their initial behavior is quite different, informed totally by their priors. However, if they live in the city long enough (see enough data), they should start to converge on similar "what should I do each day to not get robbed" behavior
I'm seriously enjoying your content, btw. I'm using Bayesian statistics in a research project of mine for grad school and your content is immensely helpful
Thanks for sharing your thoughts! Wishing you well with your project, it sounds cool 😎
Hey man love your channel and this video provided a lot of clarification for me. I think this discussion is a bit more applied than you imply. In my day to day data analysis, I'll be using the frequentist approach for the vast majority of the time, but now you have nudged my mind to consider whether a Bayesian approach might be useful in certain circumstances with limited data that makes the frequentist approach inapplicable. Basically, I can see applying a Bayesian approach in the right context being a great way to think laterally, get creative, and even generate interesting hypotheses that can possibly be tested later using a frequentist approach! This is tied to another concept you mention in this video; that of the Bayesian converging with the frequentist as data become available.
I’ve been looking all over for an explanation of the difference between frequentist and Bayesian statistics. Your video finally cleared it up for me, thank you so much for the quality explanations! Your channel has been feeding my growing interest in statistics 😄
Love to hear it! Keep on learning
Superb! I'm a "measurement uncertainty geek". You wrapped a ton of concepts into your speed of light example. Even if everyone concedes that Einstein was correct and that there is only one possible value for this constant, all attempts to measure it will produce a variety of results. Also, I belong in the camp that believes that it is a waste of time to even discuss the "true" value for this parameter because it just adds confusion and cannot ever be attained down here on Earth. All that we can reasonably expect is to narrow the range of possible values down to a point at which the possible variations is too small to bite us on the ass when we need to depend on one value for some other purpose. Thank for wandering a little distance from your normal, very practical starting point, which is so very powerful and trustworthy.
Amazing video !
A common misconception is that Bayesian priors are subjective probabilites based on personal experience and so computed results are at the whims of individual beliefs. Priors in science are almost always objective probabilities drawn from current scientific knowledge to which observed data is then added via the likelihood to generate a posterior probability. Priors are no more subjective than choosing the method for acquiring frequentist data or choosing a frequentist model.
Well said!
I love u man, I am still learning a lot from u!
Fantastic introduction to Bayesian stats!
This was quite enlightening. I have to watch and read more about it though, because my gut impression, even after the video, is that Bayesianism just feels like bullshitting prejudice when you don't have the evidence at hand.
great explanation for both point of view!
Bravo, Maestro!
Thanks for the greatest explanation !
Great video!
Awesome vid
In some instances you cannot collect data or do the experiments. Examples from the cold wars were bombers carrying nukes and nuclear subs today. Another is search patterns looking for planes or ships lost at sea.
cool explanation
great video.
Awesome video
Thanks!
I really enjoy your work. You make it look all very easy, as it is really not. I am curious what your thoughts are about epistemic and aleatory uncertainties
What's your opinion about DSGE (Dynamic Stochastic General Equilibrium) techniques used in Economics?
Great video! I'm probably totally off on this, but it seems like frequentism benefits from thinking about a multiverse, because you can imagine that events which happen once in our universe actually have an objective long-run frequency when you consider them across all universes. So when asked to give a range of possibilities for the speed of light, they might consider all copies of themselves across other universes, some of which will live in universes with a different speed of light. It seems like it might be more of a practical limitation (can't do much with a sample size of 1 even if many other samples might actually exist) than theoretical.
Based off my previous experience with this channel I have a 90% like rate....so I'm going with that....don't need to sample all of your videos and then all the other videos that would have similar content and then feel confident to know that a thumbs up is warranted. But if the speed of light didn't have a normal distribution then....
1 if by land
2 + some inferencial noise if by sea!
thanks for this important topic, i never been satisfied with statistics and propbablity either their definitions or mathematical formulation, for example what the term "density" really means ?
hey ritvik, do you think the universe is deterministic?
Great!!
Yes, I like this discussion a lot, but I think it's actually more about politics [lol].
That's not quite true. Both Bayesians and Frequentists have a notion of "a priori" probability of *events* . But they define it on different things. For Bayesians the data are fixed numbers while the model parameters are assigned probability distributions, while for the Frequentists the data themselves are samples from a population (hence sampling probability) while the parameters are fixed. *This* is the crucial difference. By treating the parameters as random variables, Bayesians can build hierarchical models, where the prior "hyperparameters" get hyperpriors (i.e. PDFs) as well. There are also various schools of thought *within* the Bayesian camp with respect to what probability is. Interpretations like de Finetti etc, having to do with exchangeability, symmetries etc.
leeets goo
came here after watching spiderman