He has subtracted weighted gini index from orig Gini index as delta, so the final value is info gain. so he is correct on splitting the attribute based on B. . Here weighted gini for A attribute = 0.34286 and weighted gini for B attribute = 0.31668 Also, if you consider this, we get the same result.
@@rahulshah3325 but info gain and weighted gini are 2 entirely different concepts. one is related to impurity and the other related to probability of being misclassified .
Hello sir, both entropy and gini are giving separate attributes to prefer for split, the entropy says information gain on entropy is more but gini says on b is more. First of all why are theu contradicting. Secondly which one is a better indicator. Thirdly what is difference, logically what they wish to signify? Are they indicating gajn in different dimensions?
But gain due to Gini and entropy suggests B and A as best split respectively.. hence my tree is completely dependent upon which gain I use .. why is it not consistent
These are two different ways to calculate the gain. So it is expected to get different results based on them. As per your question, yes, the model is dependent upon on the methods.
sir according to gini less gain should be splitting node ans should be A no B
He has subtracted weighted gini index from orig Gini index as delta, so the final value is info gain. so he is correct on splitting the attribute based on B.
.
Here weighted gini for A attribute = 0.34286
and weighted gini for B attribute = 0.31668
Also, if you consider this, we get the same result.
@@rahulshah3325 but info gain and weighted gini are 2 entirely different concepts. one is related to impurity and the other related to probability of being misclassified .
Hello sir, both entropy and gini are giving separate attributes to prefer for split, the entropy says information gain on entropy is more but gini says on b is more. First of all why are theu contradicting. Secondly which one is a better indicator. Thirdly what is difference, logically what they wish to signify? Are they indicating gajn in different dimensions?
HOW TO FIND entropy ,information gain or ginni ndex gain for continous variables
But gain due to Gini and entropy suggests B and A as best split respectively.. hence my tree is completely dependent upon which gain I use .. why is it not consistent
Yea i have that doubt too...
These are two different ways to calculate the gain. So it is expected to get different results based on them. As per your question, yes, the model is dependent upon on the methods.
For Gini Index Best Split is for the lowest value. Hence we have to choose A. Choosing B doesn't give the correct results.