Outlier detection and removal using percentile | Feature engineering tutorial python # 2
Вставка
- Опубліковано 20 лип 2024
- Outliers are unusual data points that differ significantly from rest of the samples. They can occur due to an error in data collection process or they are just legitimate data points and represent natural variation. In this video we are going to use percentile to detect and remove outliers from a dataset. We will use python pandas for this.
Code & Exercise: github.com/codebasics/py/blob...
Topics
00:00 Outliers introduction
01:44 What is percentile
03:41 Remove outliers from simple dataset
08:10 Remove outliers from complex dataset
14:59 Exercise
Do you want to learn technology from me? Check codebasics.io/ for my affordable video courses.
Feature Engineering Tutorial Python Playlist: • What is feature engine...
Website: codebasics.io/
Facebook: / codebasicshub
Twitter: / codebasicshub
Check out our premium machine learning course with 2 Industry projects: codebasics.io/courses/machine-learning-for-data-science-beginners-to-advanced
Most of UA-cam tutorials I've gone through are top notch, but yours is loads of miles away. What makes yours different is you start from the basics with a slow pace, which makes it easier to understand.
a true comment for a true teacher......
Best channel by far. You have become my DS - DA youtuber by far. You tackle one of the most important things that is understanding the concepts and translating it into how to do it. I recently realized that is far better to understand well the concepts than knowing perfectly each pandas function and how to code them fast. Also makes things much more interesting and fun because it clicks and makes sense.
Great video as always!
Thank you, I just had a DS interview and used the age example for the outlier question, thanks for your useful lessons
i was stuck in middle of a project due to messy datasets and then i came acoss your video..it helped a lot..Thanks a lot
You Teach from zero till the the end your pace is perfect and the best part is you provide exercises and resources , Thanks for helping us and teaching us
This is amazing, Thanks a lot for posting this valuable video!
This is exactly what I was looking for and particularly for outliers, Thanks a lot for this 👍
Glad it was helpful!
Thanks a lot Sir!!! This video was very helpful for me. I was doing things wrong till now. I was under impression that outliers are meant to deleted. Instead of wasting time on irrelevant data points.
max_threshold = df['price'].quantile(0.90)
print(max_threshold)
min_threshold = df['price'].quantile(0.05)
print(min_threshold)
df1 = df[(df['price'] > min_threshold) & (df['price'] < max_threshold)]
Really, I am Learning a Lot from Your Channel..Waiting everyday for your videos..Your way of Explaining Concepts, Giving Exercises and genuine Talks makes your channel Different from Others..Thank you once again
how to decide the quantile values?
You never go wrong watching codebasics, I soo much love your work sir.
👍👍 thanks for you kind words of appreciation Atunrase 🙏
It is an amazing explanation, eager to watch more videos on this topic....thanks a lot for sharing your knowledge and skills :)
I am glad it was helpful
Thank you for this video.. Got to learn a great way of how to detect Outliers and remove them
would request you , if possible please add the link to the playlist of this series in the description of the video... though i have bookmarked it.. but it becomes really easy :) thank you very much for your good work again
Thank god u again started making videos😀😀
thanks alot. i have question . should we do this procedure for all features one by one for detecting outliers and then remove it?
Great video!!Please upload more methods to detect and remove outlier detection
you have created amazing series Sir.
Thanks a lot for sharing your knowledge and skills.
thank you..'if a person cant explain simply,,means he hasnt understood/knows it properly'
following ue entire playlists
👏👏👏✔ from Brazil - Teresina - Piauí
Amazing Video... Superb Dhaval Sir🙏
☺️☺️
How about using log transformation to remove the skewness, doesn't it do a similar job with min, max thresold?
The Tutorial is just amazing, can you please magnify your screen a bit for a clearer view?
Thanks for the video .. keep up the good work.. wish you all the best
thank you
Very clear explaination! Thank you.
Glad it was helpful!
Nice explanation, thx! But what do to if my data points following are following an exponential distribution?
Amazing tutorial sir, but I have one question. How would you decide what quantile value to keep since dataset would be from different domain apart from retail price such as weather report or sales report or any other?
If I properly understood your question (good one, BTW), I would say domain knowledge mixed with an initial dataset statistical analysis.... Right?
thank you a lot great content.
Awesome sirr
great content
you just gained a new subscriber
thank you for your efforts sir
I am glad it was helpful
this tutorial was really good, Thank you
Glad it was helpful!
Hello Sir, I have been through your video and its nice. Since I am new to data analytics I have fair theoretical knowledge about quantiles
but don't know on what basis you choose the quantile level, I mean what do you look for in the describe command of any table.
please help me with this.
you are a real teacher ..... thanks....thanks a lot....
Happy to help
Really you explain it in a very easy way.
Glad you liked it
Thank you!
Tks for the great video...
Thank you dhaval bhai
thank you- really helpful
nice explanation sir like always
Sir pls upload all the videos for feature engineering.......
And you are teaching methods are great
Sure sakaham
thanks. very clearly explained.
Glad it was helpful!
Is it possible to filter out the outliers of multiple columns in a single program? please do let me know...
please make video how to setup gpu laptop. for deep learning project. i have gpu laptop but when i am starting training gpu is not process my task my haul task done on cpu. i have 4gb nvidia gtx 1650 graphics card
nice explanation. clear
I am glad you liked it
@10:24 when you take min and max quantile as .001 and .999... I am confused with 0.001... Should it not be 0.01? (meaning 1 percentile). Thank you again.
Yes, I am also. bcz 0.001 means 0.1% ?
no need to get confused he just went too low with quantile value.... happy learning.....
no, basically he excluded 0.1 % of samples in both ends (upper and lower).
the lower end is 0.1 %, which means he excluded 0.1% of samples below
the upper end is 99.9%, which means he excluded 0.1 % of samples above
thanks, this video is really helpful
Glad it was helpful!
Your videos are very easy to understand and thanks for the content.
one request from my side, volume of your videos are less compared with other sources. If possible , please increase volume
Can you check volume of your computer. I played this video on my computer and it's quite good in terms of volume
Great video
Thank you Sir
I'd like to know why you've used the variable price per area (which is just the quotient between price and area) to do the outlier treatment instead of using the mahalanobis distance (that infact, take into account the correlation between the variables) since you have 4 numerical columns.
I hope someone can answer to me. I'm new in this and I have so many questions about it :)
Hi, anyone did the Exercise? Actually, the main quantile outlier removal is done using the process but what to do with 10k NAN values in "last_review" and 'reviews_per_month" column? Apart from this exercise if we encounter that big number as NAN what should we do? Any suggestions...
How to find that where is the outlier present, because there are so many variable presents in the data set
what if data has no outlier. In that case we will loose tiny data? how to know if not outlier removal is needed in big dataset?
"quantile(0.001 , 0.999)". When we choose 0.001 or why dot not we choose 0.005 or other? Are there any regulation/ suggestions to choose these numbers? Thank you teacher!
Same question
This is a matter of intuition
Correct me if I am wrong.... We can use percentile based outlier detection only if we know that the variation is normal distribution. For the height example taken by you, we know that it follows normal distribution, so the percentile based outlier detection can be applied. What if the variable distribution is Weibull type. Can we use percentile based outlier detection?
thank you
This is good only for small data sets . But if we have big datasets with multiple of col so it's very hard to apply
Can we use this method with larger Data Set?
thanks a lot for sharing your knowledge and skills , can you please give us dataset. Thanks in advance
What if you have several variables to fix outliers for in one data set. For example What if you wanted to remove outliers in the ‘price’ and ‘price-per-square’ variables.
Sir,
How do we decide based on, which column to remove the outliers what is the logic for finding that column? I mean why cant we use other fields like total_sqft
after I got describe of data, I saw 75% of samples have total_sqft less than 1672, and max total_sqft is 52272, I wanted remove row have maximum total_sqft, is it wrong ? the way I thought
how can we fix the thresole value
What if we want to treat the outliers rather than removing them.? Which is a better practise.?
You can treat them based on a situation either of them is good
Are there outliers present in categorical data?
Sir, how to get jupiter note book?
Just one correction @12:15 you mentioned minimum threshold is 1% but isn't it actually 0.1% ?
Shouldn’t it be 97.5 and 2.5 as 95% of values are within 2SD
But how do you know if the outlier value is above 0.95th percentile?
Sir , hdbscan outlier detection pe bhe video banaye please
8:10 After this if u check outliers in box plot u will find again because of q1 q2 q3 again assigning so that's reason u can't go for remove better go for replacement
I am confused between the percentage you have taken in min and max thersold. You said that you used 1% as min but you used 0.001 which 0.1 % and 99 % as max but you used 0.999 which is 99.9%. Plz clarify this issue sir. Thanks for this hard work for us.
I actually meant 0.1% and 99.9%, exact percentage or quartile can really vary on case to case bases. You basically use your sense of judgement to come up with these threshold values
where i can find 'bhp.csv file
I downloaded data from.kagle to.do hand on but when I import csv file in jupyter notebook by its name , it warn an error( name error, file not found
Pls help me in solving this issue.)
Attribute error :- model panda has no attribute 'read'
@@panduenglishacademy7856 df = pd.read_csv('yourfilename.csv') . it should be like this format.
Why quantile? and how quantile works?
Is Outlier and Imbalanced are same concept or different?
if different could you please share some information...
i tried to find based on Definition both seems to be same, but both have different methods to detection and removal. So bit confused.
They are different concepts. By imbalanced most likely you are referring to imbalanced data sets in terms of machine learning where one class label have very less samples compared to another class label
@@codebasics - Does it means that Imbalanced data is mainly for Target / Class label and Outlier is for other features from the dataset?
There are different ways to remove outliers, when to use what??
can yo do an example of removing outliers using box plot
Yes that is coming up
@@codebasics thank you very much
Sir i am not able to download this csv file, it says that data is very large, what to do?
Can you just git clone it?
@@codebasics yes sir i did, but it says we can't make it to csv ( only raw data is available)
This wont work in case if the data contains na values
Good explanation mate ! However you can apply fix percentile removal only on toy datasets.
For real world data you would Hampel filters, Isolation forest, rolling window based MAD etc for outlier removal.
Sir secand wala dataset chahiye
the outliers in this video are the mean prices today 😅
min_thresold = df['price'].quantile(0.01)
_thresold = df['price'].quantile(0.9999)
Data set please sir
You can find it on GitHub page
exercise done
github.com/styagi9817/oulier-detection-using-quantile-funtion
That’s the way to go Shubham, good job working on that exercise
who else spot binod at 4:09 lol ;()
binod
4:00
Binod xD
Thank you so much sir