Decision Trees

[@nad4153]

Man I don't know how to thank you for all you videos. You make knowledge clear, accessible and free. All heros don't wear cape

[@ritvikmath]

Aw :)

[@Helena-rq1qx]

Thank you for such a clear explanation! I appreciate that you break down the calculations in an easy to digest and intuitive way

[@t_geek9211]

I like your explanation. It's very cool and a little bit fishy. I've noticed that there are 3 kinds of explantions for decision trees: some people focus on entropy and information gain, others focus more on reducing impurity and gini coefficient and in this video I found third possible explanation of decision tree. It always amazes me how can we end up with the same solution using different methods. Thank you!

[@ritvikmath]

Good observations! Indeed, there are always many avenues to understanding.

[@vivienvuong1218]

This is so, so helpful. Thank you so much for sharing this!

[@mamahuhu_one]

Great video! Do you have a reference explaining this and further?

[@SuperMtheory]

Great video. Could you follow up by creating a similar example that isn't as symmetric? Also, are you using Bayes' Theorem?

[@TheLameFlameYT]

Wouldn't we get a higher accuracy if we were just to say that if a fish is short it's a Salmon and if it's long it's a Tuna? That should give us a 75% accuracy, right? (80% in the case we're splitting by weight)

[@olistiktok]

For the (25/100)(25/100) reasonning, do you actually mean that it comes from the 25Tvs.75T (= "knowing that it is short) times the 25Tvs.75S (prob that it is a thuna and not a salmon) ? The added symetry in the quantities you chose confuses me i admit..😅 thx

[@Fat_Cat_Fly]

so great!!! thanks!!! love your video!!!

[@ritvikmath]

Glad you enjoyed it!

[@chenqu773]

Great tutorial. However I still coun't get how you did the conditional probability calc using that formula. I knew only that famous formula for conditional probability as P(A|B) = P(AB)/P(B) but you did not use that.

[@thomasc7526]

This is using the Law of Total Probability

[@jayantachakraborty4915]

I didn't really understand why its (25/100)(25/100) and similarly (75/100)(75/100). Can someone clarify?

[@chenqu773]

I think finally I figured it out. Taking (25/100)(25/100) for example. This literaly answers: Given that the fish is short (and with only this information), what is the probability to make a correct classification (or, a correct guess). So, if you randomly pick a fish and claims that it is tuna, what have you done? You have: 1. Picked a fish. 2. Claimed that the fish is a tuna. Both 1 and 2 is subject to a probability of 25%, so you have a chance of correctness of 25%*25% if you randomly pick a fish and claim that the fish is tuna. For the same reason, you have the chance of correctness of 75%*75% if you randomly pick a fish and claim that the fish is samon.

[@bungaIowbill]

It doesn't change the main point, but there's no real reason to randomize the predictions within the nodes -- if there are 75 salmon and 25 tuna in the node, just guessing salmon is preferable. This gives a probability of being correct of 3/4, which is better than 5/8.

[@davidporterrealestate]

excellent