Gini Index and Entropy|Gini Index and Information gain in Decision Tree|Decision tree splitting rule
HTML-код
- Опубликовано: 13 янв 2020
- Gini Index and Entropy|Gini Index and Information gain in Decision Tree|Decision tree splitting rule
#GiniIndex #Entropy #DecisionTrees #UnfoldDataScience
Hi,
My name is Aman and I am a data scientist.
About this video:
How does a Decision Tree Work? A Decision Tree recursively splits training data into subsets based on the value of a single attribute. Splitting stops when every subset is pure (all elements belong to a single class)
This video explains Gini and Entropy with example.
Below questions are answered in this video:
1.What is Gini Index?
2.What is Information gain?
3.What is Entropy?
4.What is tree splitting criteria?
5.How is decision tree splitted?
About Unfold Data science: This channel is to help people understand basics of data science through simple examples in easy way. Anybody without having prior knowledge of computer programming or statistics or machine learning and artificial intelligence can get an understanding of data science at high level through this channel. The videos uploaded will not be very technical in nature and hence it can be easily grasped by viewers from different background as well.
Join Facebook group :
groups/41022...
Follow on medium : / amanrai77
Follow on quora: www.quora.com/profile/Aman-Ku...
Follow on twitter : @unfoldds
Get connected on LinkedIn : / aman-kumar-b4881440
Follow on Instagram : unfolddatascience
Watch Introduction to Data Science full playlist here : • Data Science In 15 Min...
Watch python for data science playlist here:
• Python Basics For Data...
Watch statistics and mathematics playlist here :
• Measures of Central Te...
Watch End to End Implementation of a simple machine learning model in Python here:
• How Does Machine Learn...
Have question for me? Ask me here : docs.google.com/forms/d/1ccgl...
There is a mistake in your video:
You said to choose that attribute that has less information gain. But actually we have to choose that has high information gain...
Yes Naat, thanks for pointing out. I have pinned the comments related to it in the video for everyones benefit.
@@UnfoldDataScience Pleasure sir
If you are saying that we have to choose high information gain. Then as per video we should take the impure node. For pure node gini would come 0 and hence 0 IG. Isn't something wrong.
At what time that has been said and corrected?
@@DK-il7ql 10:37 he said Low information gain by mistake instead of high information gain
I usually don't like commenting on RUclips videos. But for this one, I felt like I had to show appreciation because truly this video was extremely helpful. University professors spend hours explaining what you just explained in 11 minutes. And you are the winner. Perfect explanation.
Thank you so much!!!!
I appreciate it Ahmed. Your comments motivate me :)
Institutes spend two hours in explaining these two concepts and you made it clear in some minutes.excellent Explanation .
Thanks a lot :)
@@UnfoldDataScience I agree
This has become my favourite channel for ML/Data Science topics,thank you very much for sharing your knowledge
Thanks Jehan, your words are my motivation.
Wow! Not only was your explanation amazing but you also answered every single comment! True dedication. Keep it up!
Thanks a ton Zain.
I have an assignment due tomorrow and this helped a lot!
Excellent, to the point, good examples. Great work!
Best channel to learn ML and Data science concepts. Thank you sir
Thanks Indrajit. Kindly share video within data science groups if possible.
If i feel any concept is hard to understand, first thing i do is search for your videos. Very intuitive and easy to understand. Thank you so much!
Your comments are my motivation Akhil. Thanks a lot. Happy learning. Tc
The simplest and best explanation so far.
Glad it was helpful Shyam.
I just discovered this channel what a gem
Thanks a lot. please share with others in various data science
groups as well.
finally i am getting some clear explanations for various concepts
thanks Indra.
You have a really good explanation skills, thank you man , i finally understand it
Most Welcome :)
short simple and sweet, thank you so much
You're welcome Kunal.
Thanks bro...explained in easy manner...
thanks for clear and easy explanation
With your clear explanation, I finally understand what Gini index is. Thank you so much!
You are welcome. happy learning. Stay Safe!!
This is brilliant. Thank you so much!
Thanks Abhijit. Keep Watching. Stay Safe!!
Very goog and clear, i'm french speaking and i had understood almost everything
Thanks Hassan.
Thanks for the video ! It was really clear and well executed. Would have been great to detail the entropy calculation though, I find it a bit elusive without an example
Simple & clear
Thanks a lot.
This is On point, thank you so much.
You are so welcome Bhargav.
Thank you for this video very helpful
Your first video that I came across. Subscribed!
Thanks Vishesh.
I appreciate your concepts for Gini and Entropy
Thanks Awanish.
love it, very clear explanation
Thanks Eider. Happy learning. Tc
Crystal Clear Sir!! Keep Going!!
Thank you Anandram.
Amazing explanation sir
thank you so much, this helps me a lot!!!
I'm so glad!
very clear explanation and very helpfull
Glad it was helpful Deepika.
Great Explanation !! very helpful . Thank you :)
Glad it was helpful!
Thanks for this clear and well explain Gini index.... Thanks ....
Glad it was helpful!
sir Your explanation really very much helps me thank you
You are welcome.
Awesome video.. Thank You so much!
Thank you.
Your explanation is awesome, thanks.
Thanks a lot for your valuable feedback.
Thank you so much sir, before watching this video I have watched 4 videos related to impurity but everyone is doing mixup of entropy and impurity n it was not really clear like what exactly formula is, how does it works.. But after watching ur video.. It is tottaly cleared now. Thank you for this beautiful n clear explanation
Glad you understood
Amazing explanation
Thanks Ravi.
U r doing great job sir
Thanks a lot.
Thank you, well explained
Glad it was helpful!
Thank you, sir!
Very welcome!
thank you so much
Thank you, no one could have done better
You comments mean a lot to me
Thanks man
Thnaks alot
Welcome
It was very informative, Sir. Thank you :)
Most welcome Prerna.
great explanation
Glad it was helpful!
Great video
Thanaks a lot.
Thank you
Welcome Jarrell.
Hello Aman,
Hope you are well. I have a question. Hope you can help me here.
If probability(P) =0,
Then Gini Impurity becomes = 1,
as per the formula.. Then why it always ranges from 0 to 0.5?
Thank you,
Subhajit
Great video !
Thanks for the visit
Very nice explanation and icing on the cake for comparing their performance at the end.
Just to confirm, is Gini/IG only for classification?
For the regression trees we would use loss functions like sum of squared residuals?
That's a good question, since it's based on probability so it is applicable to classifiers. For regression, we see something like to minimize SSE or other error.
@@UnfoldDataScience
Hi sir, as per my knowledge "Information Gain" is used when the attributes are categorical in nature. while "Gini Index" is used when attributes are continuous in nature.
Hi sir, as per my knowledge "Information Gain" is used when the attributes are categorical in nature. while "Gini Index" is used when attributes are continuous in nature
Thank you for your wonderful explanation.
Please make a video on PSI and KS index.
Will do soon Sager. Thanks for feedback.
Thanks a lot!
You're welcome!
Good teaching
Keep watching
Great content.
Thank you.
Nicely explained....! Subscribed :)
Thanks Lalit. So nice of you :)
i have one question aaman . at root node is the gini are Entropy is high are low..
Which one to choose, like how by seeing the data I can assume, what we can use gini or IG?
Cant decide in advance, its more of trial and error(there are some directions though)
very well explained
Thanks for watching Subhangi.
does the CART go through all the possible numerical values under loan to find the best condition? If you have a large amount of data, then should it be very slow?
That is a good question. Thanks for asking. In general, for a numerical variable, first split point is chosen randomly and then the point is optimized based on "in which direction" loss function is moving. Please note, loss in this case is the node purity after split.
Thank you ❣️
Welcome.
Very Good!!!
Thank you Chris. happy learning. stay safe. tc
Can you help to explain intuitively the Entropy equation
So, the only difference between Gini and Information Gain is only the performance speed right? I assume with the same state of descision making and data, both Gini and Information Gain will be able to pick the same best attribute, right?
Great video btw!
That is correct. Also the internal mathematical formula is different.
Thankyou sir
Welcome Siva.
Amazing explanation aman . I have one doubt like suppose there are 5 columns(4 independent and 1 target). For split i have used 1,2,4,3 columns and other person is using 3,2,1,4. Then on what factors we can decide either my splits are best or the other guy's split is best.
Its algorithm decision which columns to use.
Awesome work and very intuitive explanation! Thank you. I have an exam in Data Mining and you helped me sir!!
Glad it helped! Happy Learning!
thanks
Welcome.
Hi Great explanation. Thank you so much. Do you have any videos explaining the criteria for Decision Tree regression?
Thanks a lot. for Regression, not yet, will upload soon.
So if i am using the C5.0 algorithm? Which Separation technique will be used?
entropy for measuring purity.
Great explaination
I have que
Is gini index negative
Hi, no it can not be.
Just to make clear, the Gini index ranges from 0 to 0.5 and not 0 to 1. Jump to to video at 7:10
Yes, this is the common comment from many users. Your are right Abhishek.
sir how you choose the loan amnt as root node ?we have to find gini for all columns and then select the root node?
Hi Amna, this is a good question. Thanks for asking. Yes for all olumns and
select the optimal split.
Do we have to calculate both gini and entropy to figure out which is best for the dat
aset??
Only one at a time.
Firstly sir , how much i know higher the information gain gooder the split.
& I wann know that is any of them is for continues variable?
Higher IG is better
Thank you so much sir please do some projects
Thanks Vishal.
well sir how the root node selection criteria occurs if two data sets shares same and lowest gini index value
Happens very rarely, Geethanjali.
But if we have datasets with multiple columns like more than this example then how we will decide select which input column should be splited?
Answered.
Hi! i want to make sure about gini index. You said that "criteria of the split will be selected based on minimum GINI INDEX from all the possible condition". Is it "gini index" or "weighted gini index"? Thanks a lot tho! Learn a lot from this video!
Thanks Steven. "Gini Index".
Good explanation.. But correction needed. Gini oscillates between 0 and 0.5.. The worst split could half positive half negative.. Gini impurity for that wing is 0.5 also overall weighted gini would be 0.5..
It is entropy that oscillates between 0 and 1.
You are Right Anil. This feedback is coming from other viewers as well may be I mentioned this part wrong in video. I am pinning your comment to top for everyone's benefit. Thanks again.
I'm a bit confused between Gini and Entropy. I mean is it necessary to use both methods while analyzing or we can go for any one of them?
We have to use only one of them. Which one to choose depends on data.
Depends on case not both to be used
✌🏻✌🏻
which model has less bias and high variance-logistic, decision tree or random forest? can you please help
Decision tree high variance low bias
Logistics regression - high bias, low variance
Random forest - Tries to reduce the high variance of decision tree. Bias is low.
@@UnfoldDataScience Thank you very much.. can you also share the reason behind this.. or if you got any link where i can understand
Awesome Explanation, very sharp! I have 2 questions:
1. Since this algorithm calculates Gini index for ALL splits in EACH column, is this process time-consuming?
2. What if the algorithm finds TWO conditions where GINI Index is 0. Then how does it decide which condition to split on?
Thank you in advance!
1. It is process consuming but it does not happen one by one internally for numerical columns, algorithm tries to figure out in which direction it should move smartly. For categorical columns it happens one by one and time consuming.
2.0 means homogeneous sets hence no further split will happen
I calculated the Gini Index for (4, 2) splits, it came as 4/9. Shouldn't it come close to 1 ? Since it is the worst case scenario?
Need to check with data and calculate however not always mandatory that it will be close to 1.
it is a nice tutor Sir ! But how could it be such category comes true ? since you made greater or equal to 200 and should be inclusive to the GINI index ?
Yes, that mistake I accepted already 🙂
Can you show one numerical example using entropy? when the formula starts with a negative sign, how can the value be positive? Just curious.
because log(x)
@10:38 where the information gain is high ,there we try to split to node right??
That is a good question. The formula you see @10:38 is for entropy of a node.
Information gain for a split = Entropy of node - Entropy of child nodes after the split
Decision tree splits at the place where the information gain is highest. In other way you can say , decision tree splits where entropy is reduced to largest extent.
Sir kindly explain entropy in detail just like the way you presented gini index
Sure Karthik. Keep watching.
Sir, range of Gini Index is from 0 to 1 or 0 to 0.5? i am confused
see previous comments. we have discussed it.
How gini index ranges from 0 to 1? For best case it is 0 and for worst case it is 0.5..then how it is possible? Please explain..
The coefficient ranges from 0 (or 0%) to 1 (or 100%),
I think we select the split with the highest information gain when using entropy. Please correct me if I'm wrong.
You are right, When an internal node is split, the split is performed in such a way so that information gain is maximized.
Thanks Abdo. Yes maximum IG is considered for split. Probably I missed to include in video.
@@UnfoldDataScience You are welcome. i also get some new informations from your video
For titanic dataset what type of criteria we have to use??
Hi Noman, cant say, we need to try and see which one works better.
hello, very insightful. You almost explained the best times to use either of the criterion. Can you shed more light into that. The best kind of criterion to use for data in a model
Hi Anthony, it is usually not easy to say which method(gini/entropy) works on what kind of data beforehand. Usually we try to check with various options to see model performance and then choose one. Hope this clarifies. Thank you.
@@UnfoldDataScience Yeah Thank you.
can i get your email? I'd like to stay in touch
Sure it's there in my RUclips.
Then which splitting criteria we should use Gini or Entropy i. e information gain?
Depends on Data
Wouldn't Id two loan amount true will be on the side of true?
Yes if it satisfies the tree splitting criteria. I will have to check once. please check the splitting criteria again. Thanks you. Happy Learning!
Aman, Can you please explain entropy also with an example like you did for Gini Index
Yes Prasanth, I will try to cover that topic in one of the upcoming video.
Thank you Aman
When shoud we use gini index and when should we use entropy?
It's a question of tuning parameter based on your data, for some input one. Ay work better for some other one.
sir, i think id is not useful attribute to make model prediction. so we can get rid of that
Absolutely, these comments give me motivation that my viewers are connecting with me. Here id is just for showcase purpose. Id does have have any meaning in data science models.It should be removed when we fit the model. Thank you. Happy learning.tc
Sir why decision tree gives good accuracy in imbalanced dataset compared to logistic regression
Good question Ashish, It is because, there is no mathematical equation involved in Decision tree hence learning happens purely on rules.