In common conversation we use probability and likelihood interchangeably. However, statisticians make a clear distinction that is important to understand if you want to follow their logic. The good news is that they are both super simple. The bad news is that they are easy to get mixed up. This StatQuest gives you visual images that make them both easy to remember so you'll always keep them straight. NOTE: This video was originally made as a follow up to an overview of Maximum Likelihood kzbin.info/www/bejne/jpbTiaeibr5-rcU . That video provides context that gives this video more meaning. So if you are wondering what a likelihood of 0.21 represents, please watch that video.

Thank you!

Thanks!

Thank you very much! :)

Thanks!

There are lots of ways to create embeddings - and this video describes those ways. However, BERT is probably the most commonly used way to make embeddings with an LLM.

Thanks!

Thanks!

bam! :)

:)

Thank you!

:) nope!

:)

Glad to hear it's still helpful! :)

That would be awesome!

TRIPLE BAM!!! Thank you so much for supporting StatQuest! :)

:)

Thanks again!

Thanks!

Thanks!

Thank you!

:)

The residuals are the vertical distances from the line to the data points. We then square those distances.

@@statquest got that!

Because the process of bootstrapping requires random sampling, you'd get something similar, but not the same, as what I have. That said, I believe the "shifted" values are -4, -3.5, -2.2, 1.2, 1.5, 1.6, 2.5, 3

@@statquest Many thanks! When I try to reproduce your results, I get close on the p value of 63%, but I'm way off from you in the split. I get about even percentages in the tails, like 32% 29% -- I'll keep working on it ;-)

@@stuartkramer1082 Here's my R code: raw.data <- c(-3.5, -3, -1.7, 1.7, 2, 2.1, 3, 3.5) mean(raw.data) raw.data - mean(raw.data) shifted.data <- c(-4, -3.5, -2.2, 1.2, 1.5, 1.6, 2.5, 3) mean(shifted.data) data.1 <- c(-4, -4, -2.2, -2.2, 1.2, 1.5, 2.5, 3) mean(data.1) set.seed(42) the.means <- c() for(i in 1:1000) { boot.data <- sample(raw.data, size=length(raw.data), replace=TRUE) the.means <- c(the.means, mean(boot.data)) } hist(the.means, breaks=20, xlim=c(-4,4)) sum(the.means < -0.5) sum(the.means < -0.5)/1000 sum(the.means > 0.5) sum(the.means > 0.5)/1000 sum((the.means > -0.5) & (the.means < 0.5))/1000 the.means[the.means > 2.4]

@@statquest Well, now I'm confused ha ha ha. When I plug in data.1 into my excel variation of your model (-4, -4, -2.2, -2.2, 1.2, 1.5, 2.5, 3), I get the result you show in the video. So, that's great. And I see how raw.data turns into shifted.data consistent with your explanation in the video as to how to derive the null hypothesis distribution from the initial sample. My confusion is figuring how data.1 (the average of data.1 is about -.5) is derived from shifted.data, cause data.1seems to be the basis of the null hypothesis graph starting at around 3:30 in the video.

@@stuartkramer1082 Ooops! It looks like I messed up when calculating the numbers. I should have used "shifted.data" in the loop instead of "raw.data". That said, when corrected, the p-value is still >0.05 so we don't need to change the result.

You just add all of the samples back to the pool and bootstrap from the original dataset (with updated weights).

@statquest thats what I'm confused about, how do you update the weights in the main dataset, there will be some data samples which would not be in the current bootstrapped dataset, how to deal with their weights?

@@hammadrehman5753 You just use whatever their old weights were and then do the normalization step.

Thank you very much!

Thank you! Cheers!

Thank you!

You made me laugh out loud.