Watching MIT OpenCourseWare videos identifies how completely lacking in substance my college education really was.

*My takeaways:* 1. Nearest neighbours 4:18 2. K-nearest neighbours 8:11 3. Performance metrics 16:50 4. logistic regression 30:20

An unintended consequence of learning the difference between linear and logistic regression from Prof. Guttag in this video: the scene from The Princess Bride intrusively popped in my head where Miracle Max says: "There is a big difference between mostly dead and all dead. Mostly dead is still alive." Then I spent a few minutes giggling before I can focus again.

Listening to my professor try to lecture on classification makes me nauseous and hate my life. Listening to this guy lecture on classification and I'm actually enjoying it and understanding it. People are not equal.

I LOVE this man. I found this video from a search and didn't see the other 12 videos in the series and I am picking up everything he is saying. Also, it's so cool how he encourages class participation. Great teachers are amazing and a gift to humanity.

Everythime I see something like that, I wonder how brilliant a human can be to break down complexity so simple like that.

Prof. Guttag and Grimson are really great ! I wish I had those professors when I was in college !

Precission is Positive predictive value and not specificity! 19:20

Absolutely the best ML courses I have encountered and I have tried many. This is the only one that keeps me focus and intrigued :) Do get Prof. Guttag's book! Good companion for this class

this man seems like an old very wise man that spent most of his time either in the research, or with his grand childs plus he's a good teacher

Hats off to you sir for the way you share you knowledge.

Amazing lecture as always! Thanks for great resources👏

Stopped at the most important point 😀 makes you go to the next lesson . Anyway I am glad he mentioned it

42:46 line "for i in range(len(probs)):" because you just need i as a index for testset and probs, you could zip these lists, e.g. "for p_i, ts_i in zip(probs, testSet):" then you can use p_i and ts_i instead of probs[i] and testSet[i]. All in all a really good lecture, thank you very much!

34:24 fiting linear regression into a range, logistic regression. The machine learning model that's always visualized.

Thanks a lot for this classes, not only learn about the computational thinking, also the reason that Leonardo DiCaprio dont survive in the movie.

YES this is what I need to do the titanic challenge on kaggle

very nice this videos about machine learning

Prof. Guttag talks about problems finding k nearest neighbor for large data sets due to number of distances needed to be calculated. Good old-fashioned relational databases have had a solution to this for decades. They use, for example, partitioning, multi-level indexing and calculated columns. The calculated columns can be stored or cached. In fact, we want our database to live in a disk / cache balance that optimizes our multitude of parameters , which boil down to preprocessing time and real-time processing time as constrained by money. This makes finding nearest neighbor, or any other math based comparison, faster by multiple orders of magnitude for large data sets. Recognizing, much of this can be done in memory, my question is, at which key places in machine learning do we most apply what we have learned in other data science fields about quick data access? In other words, where can we largely mitigate these and how do we decide if it is preferable to maximize performance of a function as opposed to utilizing a different ML approach?

This is complicated, but the prof explains things well. Now i just need to learn more python to be able to understand it fully...

Can anyone here tell me where can I find the video for "Regression" from the same set of MIT videos? Under what name is it present in the MIT lecture videos from the above-mentioned link?

27:30 Wouldn't it be better to set label and k as keyword arguments instead of creating a separate knn function via lambda abstraction? He talks about using this to build much more general programs, yet he created two functions when you could just create one that does both, which would be more general than creating two.

Why didn't the professor fall in the 'dummy variable' trap? He used classes C1, C2 and C3 but he shouldn't have used all the 3 to create the regression model since C1 = 1 - C2 - C3, which means that the variables are dependent on each other. Someone knows the answer?

Thank you.

12:01 why is the k nearest neighbor data training separated into testing and training again?

Lo máximo!

3:12 I believe that this statement is wrong. He is ACTUALLY using the full representation using the number of legs too. If you do the math, using the binary rep only, then the distance matrix shown is incorrect. CORRECTION: They are NOT using the number of legs, however, they erroneously threw a 2 in the last element of the binary data for chicken while it should be 0. I tested my own algorithm with this number and I get the same result as shown in the video. Additionally, the last binary feature should be 'reptile', correct? In the python data set the last element is zero in various of the reptile cases. Please let me know if I am missing something obvious...

i'm already sad that im gonna finish these series of lectures soon :/

wow, i wish you were my professor

"The more legs an animal has, the less likely it is to be a reptile."

it turns out that linear regression and logistic regression use the term coeff to denote weight. that interesting

Is there any way to get access to the posted code?

Finally got applause for something 😂😂😂

Wow I'intersting it

tough crowd here..

If this a level of MIT, forget it!! There are much more usable courses for example professor Gilbert Strang.