input stringlengths 0 513 | output stringlengths 0 513 |
|---|---|
a constant drift problem, which is also kind of hard. This is sort of | I don't know if you're trying to measure like the height of the Rocky Mountains. |
I don't know if you're trying to measure like the height of the Rocky Mountains. | There's a constant drift for them. They're all going up by like an inch every year, or something like that. So if you're trying to do that over time, it's gonna completely change |
There's a constant drift for them. They're all going up by like an inch every year, or something like that. So if you're trying to do that over time, it's gonna completely change | and sometimes there's a drift that occurred at a specific time. So you can kind of key in on that data. Maybe remove it. |
and sometimes there's a drift that occurred at a specific time. So you can kind of key in on that data. Maybe remove it. | Maybe you can use some of the methods we talked about for normalization to actually normalize that data in a clever way. So it can still be used in the data set. |
Maybe you can use some of the methods we talked about for normalization to actually normalize that data in a clever way. So it can still be used in the data set. | Okay. |
Okay. | so we covered a bunch of stuff on normalization data, balancing and drift. |
so we covered a bunch of stuff on normalization data, balancing and drift. | And if you want some more details to dig into some of these methods. |
And if you want some more details to dig into some of these methods. | and that'll be it for today. And then on Wednesday we'll start to talk about hypothesis testing. So just be prepared for a bit more statistics on Wednesday. |
and that'll be it for today. And then on Wednesday we'll start to talk about hypothesis testing. So just be prepared for a bit more statistics on Wednesday. | But it'll be important for us to basically understand these questions of is this distribution, this other distribution? We use hypothesis testing for that alright thanks. Everybody. |
But it'll be important for us to basically understand these questions of is this distribution, this other distribution? We use hypothesis testing for that alright thanks. Everybody. | Take a little bit of a break, and then David will be showing up in a bit to talk more about the projects and the milestone. |
Take a little bit of a break, and then David will be showing up in a bit to talk more about the projects and the milestone. | |
Hmm. | |
Hmm. | alright, everybody. Let's get started |
alright, everybody. Let's get started | definitely some stuff to cover for today. If you remember, we're talking about super engineering. |
definitely some stuff to cover for today. If you remember, we're talking about super engineering. | And we got around the area being able to talk about outliers |
And we got around the area being able to talk about outliers | so definitely wanted to start with questions here, to start out like understanding how to use what are basically blocks and whisker plots to find out buyers. |
so definitely wanted to start with questions here, to start out like understanding how to use what are basically blocks and whisker plots to find out buyers. | Okay. |
Okay. | you can definitely do this |
you can definitely do this | by the Sunday. |
by the Sunday. | There's different ways to do this. You can also do it like for your whole data set. But sometimes it's good just to look at individual dimensions so you can box them. Wizard plot. Every feature, start to look for some really large outliers inside of the feature space. |
There's different ways to do this. You can also do it like for your whole data set. But sometimes it's good just to look at individual dimensions so you can box them. Wizard plot. Every feature, start to look for some really large outliers inside of the feature space. | Okay. |
Okay. | this is one of the like, st tools you can use. It's a nice way to kind of bin some of the things that look more obviously like |
this is one of the like, st tools you can use. It's a nice way to kind of bin some of the things that look more obviously like | outliers, because they'll be outside of these points. And when you do the box box and whisker flop |
outliers, because they'll be outside of these points. And when you do the box box and whisker flop | and something like python. They'll give you these different things, too. So you can end up finding out with one of these scenarios. |
and something like python. They'll give you these different things, too. So you can end up finding out with one of these scenarios. | Okay? |
Okay? | So so that was starting out with boxes. Lister plots. You can also, just, you know, use your amazing human abilities just to be able to plot everything in a scatterplot and start to see if they are actually outliers, and then you can go on and look at those later. |
So so that was starting out with boxes. Lister plots. You can also, just, you know, use your amazing human abilities just to be able to plot everything in a scatterplot and start to see if they are actually outliers, and then you can go on and look at those later. | This is when it's not so easy to understand if something is like |
This is when it's not so easy to understand if something is like | statistically different from something else, just by looking at a box and whisker plot. |
statistically different from something else, just by looking at a box and whisker plot. | So even if we start to look here like there does. |
So even if we start to look here like there does. | I would hope some of us would think there's probably some sort of pattern in the data |
I would hope some of us would think there's probably some sort of pattern in the data | that exists here. This is where most of the statistics are. |
that exists here. This is where most of the statistics are. | Maybe these can be included. |
Maybe these can be included. | Whether or not we have to investigate them. |
Whether or not we have to investigate them. | These ones, though, don't look like they're very well aligned with the rest of the data. So we can do a little bit of visual inspection. |
These ones, though, don't look like they're very well aligned with the rest of the data. So we can do a little bit of visual inspection. | This is really more in the case when you don't have a lot of data. Yeah, you can try and use your human abilities to sort through a few things. |
This is really more in the case when you don't have a lot of data. Yeah, you can try and use your human abilities to sort through a few things. | But the more typical way of doing this is to basically look at finding some outliers based on your standard deviation. |
But the more typical way of doing this is to basically look at finding some outliers based on your standard deviation. | So you can compute the standard deviation |
So you can compute the standard deviation | which we've talked about a bit already |
which we've talked about a bit already | and look for data points that are, you know. |
and look for data points that are, you know. | usually between and times away standard deviations away from the meeting. This means, you know, you're |
usually between and times away standard deviations away from the meeting. This means, you know, you're | when you're times away. What I remember, you're like |
when you're times away. What I remember, you're like | in the most out there, % of the data that actually exists that you collected. So somewhere between these statistics of doing and standard deviations means that you're you're really rare. |
in the most out there, % of the data that actually exists that you collected. So somewhere between these statistics of doing and standard deviations means that you're you're really rare. | So if you're really rare, maybe you're really awesome data you want to focus on, or maybe your data. That particular point is quite crazy and can be ignored. This could be a really. |
So if you're really rare, maybe you're really awesome data you want to focus on, or maybe your data. That particular point is quite crazy and can be ignored. This could be a really. | you could just basically do something somewhat as simple as if you've got your data, find the mean, take the standard deviation, multiply it by some factor. |
you could just basically do something somewhat as simple as if you've got your data, find the mean, take the standard deviation, multiply it by some factor. | This will give you all the stuff in this like really upper limit bin and lower limit bin, and then |
This will give you all the stuff in this like really upper limit bin and lower limit bin, and then | use that data. How you want. You probably don't want to use it, but you can at least investigate it. |
use that data. How you want. You probably don't want to use it, but you can at least investigate it. | Does that make sense? |
Does that make sense? | This is one of the like spectlar things to start with for being able to check for outliers. |
This is one of the like spectlar things to start with for being able to check for outliers. | And like, I'm saying, these are all kind of like rules of thumb. They're not always perfect, depending on how your data is distributed. Maybe it is like perfectly Gaussian, and you'll still find stuff in these bins, and they're really not far away from the mean. So you should. If you're really desperate to keep the da... |
And like, I'm saying, these are all kind of like rules of thumb. They're not always perfect, depending on how your data is distributed. Maybe it is like perfectly Gaussian, and you'll still find stuff in these bins, and they're really not far away from the mean. So you should. If you're really desperate to keep the da... | Okay. |
Okay. | so this is also called like a zed store way of being able to compute outliers. |
so this is also called like a zed store way of being able to compute outliers. | It's just this sign number, which is basically a number of standard deviations away from the value for a mean |
It's just this sign number, which is basically a number of standard deviations away from the value for a mean | and this is just coming from the Wikipedia definition. Just in case you run into |
and this is just coming from the Wikipedia definition. Just in case you run into | one of the larger notation problems with data. Science is you'll go to one library is python. |
one of the larger notation problems with data. Science is you'll go to one library is python. | It'll call this like standard deviation outlier check, and you'll go to another library and this or another library are, it'll call it a Z Score check instead. So occasionally I'll have a slide like all of these terms mean the same thing. Yes, mostly because they come from different mathematical backgrounds. But if yo... |
It'll call this like standard deviation outlier check, and you'll go to another library and this or another library are, it'll call it a Z Score check instead. So occasionally I'll have a slide like all of these terms mean the same thing. Yes, mostly because they come from different mathematical backgrounds. But if yo... | okay. |
okay. | so |
so | this is similar. So if you're using this sort of Z score way of looking at this. |
this is similar. So if you're using this sort of Z score way of looking at this. | because you can still compute |
because you can still compute | by being able to rescale and center the data using a Z score. |
by being able to rescale and center the data using a Z score. | So in this case. |
So in this case. | So before, when you're checking standard deviations, you kind of just take your data, use the mean of the data |
So before, when you're checking standard deviations, you kind of just take your data, use the mean of the data | and then use your standard deviation. When you're using the Z score, what you're going to do instead is kind of rescale and center the data. So we're going to standardize the data such that basically now, it's like mean and one unit, one standard deviation. And now you can just use Z as a factor. Now that you've kind... |
and then use your standard deviation. When you're using the Z score, what you're going to do instead is kind of rescale and center the data. So we're going to standardize the data such that basically now, it's like mean and one unit, one standard deviation. And now you can just use Z as a factor. Now that you've kind... | Yeah, sorry does that part make sense? |
Yeah, sorry does that part make sense? | You're basically transforming the data. So now you can just use a number Z, which is the factor from slides ago. But just because the data has now been standardized. |
You're basically transforming the data. So now you can just use a number Z, which is the factor from slides ago. But just because the data has now been standardized. | okay, that one's occasionally useful. And if you remember, we talked a bit about this like pipeline, where you collect your data, you can use some statistics. You might standardize stuff, and then you can just use the same Z score for everything. If you keep the same pipeline that you're using for your data. |
okay, that one's occasionally useful. And if you remember, we talked a bit about this like pipeline, where you collect your data, you can use some statistics. You might standardize stuff, and then you can just use the same Z score for everything. If you keep the same pipeline that you're using for your data. | you can check for outliers inside of the standardized space instead. |
you can check for outliers inside of the standardized space instead. | Okay. |
Okay. | and this is where you know, if you take some threshold. |
and this is where you know, if you take some threshold. | I think this is was. If you can compute the Z score for everything. |
I think this is was. If you can compute the Z score for everything. | Numpy has this kind of interesting notation, which might seem kind of confusing, but you can say like where greater than some number, and it'll check stuff in the actual ring to see that the value is actually greater than this. |
Numpy has this kind of interesting notation, which might seem kind of confusing, but you can say like where greater than some number, and it'll check stuff in the actual ring to see that the value is actually greater than this. | This is |
This is | kind of weird, but you'll just have to trust me that this works inside of numpy. |
kind of weird, but you'll just have to trust me that this works inside of numpy. | Okay? |
Okay? | Yeah. The other one, which is related to the box and whisker plot is. So you can also look at interquartile ranges here. Like I said, this is also called mid spread, or the middle %, or technically H spread |
Yeah. The other one, which is related to the box and whisker plot is. So you can also look at interquartile ranges here. Like I said, this is also called mid spread, or the middle %, or technically H spread | again, different terms for basically the same thing. |
again, different terms for basically the same thing. | And it's really you want to understand the difference between like the th and or sorry the th and th percentiles |
And it's really you want to understand the difference between like the th and or sorry the th and th percentiles | and their upper quartiles. So if you had a |
and their upper quartiles. So if you had a | bunch of data. |
bunch of data. | this is supposed to be the th percentile, and this other line would be the th |
this is supposed to be the th percentile, and this other line would be the th | in here, is it like % of your data? |
in here, is it like % of your data? | And there's over here and over here. |
And there's over here and over here. | And the idea is, most of your data should be in that middle chunk. That's your kind of |
And the idea is, most of your data should be in that middle chunk. That's your kind of | normally. |
normally. | your data set seems to look like the rest of your data set. And now you just want to understand. Okay, it is. |
your data set seems to look like the rest of your data set. And now you just want to understand. Okay, it is. | There's something weird about these other chunks within the last % of the data where I will say. |
There's something weird about these other chunks within the last % of the data where I will say. | okay, the difference between the , th which is everything beyond that, and , th the same time |
okay, the difference between the , th which is everything beyond that, and , th the same time | kind of a, again, a bit of a weird notation when they're using these numbers. |
kind of a, again, a bit of a weird notation when they're using these numbers. | And this is good. This tends to be a little bit more robust to outliers. |
And this is good. This tends to be a little bit more robust to outliers. | So like one of the odd things about using the Z score from the last page. It means you do take your data |
So like one of the odd things about using the Z score from the last page. It means you do take your data | and you normalize it using standard deviation standard deviation. |
and you normalize it using standard deviation standard deviation. | Which means you've already normalized it a bit in some way, so you might have even affected a little bit about how you compute your outliers |
Which means you've already normalized it a bit in some way, so you might have even affected a little bit about how you compute your outliers | so often. This method is a in its own way. I'm a little bit |
so often. This method is a in its own way. I'm a little bit | more robust outliers because it's just gonna take |
more robust outliers because it's just gonna take | top bottom, %. Put those as potential outliers. |
top bottom, %. Put those as potential outliers. | Okay? |
Okay? | And this is where you can just use an example of this. So if you were to take |
And this is where you can just use an example of this. So if you were to take | basically the st quantile here, so you want to get the bottom % of values. |
basically the st quantile here, so you want to get the bottom % of values. | And then you can take the top |
And then you can take the top | quantile for the rest of the %. |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.