NOTE: This StatQuest is sponsored by JADBIO. Just Add Data, and their automatic machine learning algorithms will do all of the work for you. For more details, see: bit.ly/3bxtheb BAM! Support StatQuest by buying The StatQuest Illustrated Guide to Machine Learning!!! PDF - statquest.gumroad.com/l/wvtmc Paperback - www.amazon.com/dp/B09ZCKR4H6 Kindle eBook - www.amazon.com/dp/B09ZG79HXC
Really hope to see AutoML become more popular. Too many people wasting time doing things like hyperparameter optimization or such when there are much more important things to look at.
Genuinely love StatQuest! This channel has helped me though so many knowledge gaps when trying to get through my degree programme. I'm looking to assess AutoML systems for my dissertation so thanks for this video. I would love to see some practical examples on the channel - if that's at all possible or interesting to you.🙂
Hi Josh and all. Thanks for the video! Maybe could you do an autoML video series? For example, interview with people from Auto-Keras, or H2O.io auto-model companies? Thanks
HI Edwards, JADBIO does not have a video tutorial .... yet :-) , but if you want, we have a two week trial and sample data and several step-by-step tutorials that will walk you through the process of either regression, binary or multi-class classification, and survival. you can find all at JADBIO.com under our use cases.
This kind of topics always reminds me the statement by Harari in Homo Deus: _"Eventually, algorithms will be so advanced that machines will be in your position to be interviewed, only to find out the company also deployed machines to interview the applicants"_
Well, it might be in the fuuuuture, but for now. I can tell you of some companies that are testing this and it is a catastrophe. Avira, don't provide call center services anymore. When I had a problem with the VPN, the only option was an email. I sent the email and I received the answer of an AI that correctly directed me to the troubleshooting question, but the troubleshooting didn't solve the problem and the AI assumed that it did, so it sent me another email saying that the problem was successfully resolved after 1h trying to waste my time in solving the problem. As a result, I ask for a reimbursement, which I got, and I rated AVIRA VPN like trash and a company that doesn't take care of clients. That means we are not quite there.
Hey Josh, thank you so much for all these videos on your channel ! They saved my life more than once ! As a suggestion, I think you should consider making a video about Partial Least Squares (PLS) regression, it's a quite nice and efficient method, but I remember struggling so much with it back then !
Building a Machine Learning model involves lot of analysis , discussion and decision making while treating the data, choosing ML algorithm, inferencing ML model results ... Any automation can make faster ML model building by scalable infra / any other way But the irony is building a ML model is next best priority after understanding business problem, sourcing right data and treating the data as well
AutoML is what i suggest the future... like automation always is. like things we use today and think we have to do many stuff manually ... in fact... people earlier called building up the basis for that: automation
Hi Statquest. Love your content. Been following for some time. I was looking at entering the Data Science and ML field?? SHould i learn how to do hyperparameter optimization or just learn Auto ML for doing it?? Your response will be much appreciated.
It's still important to learn how to tune hyperparameters because autoML is only getting started and you might want to use a model that isn't part of it.
Hi! Thanks for creating and maintaining such a useful channel! Multiple BAM's! :) I looked for videos that clearly explain Apriori and Ripper, but couldn't find them... Did you create those? RU-vid is full of examples... but no one else does it like you do. (Thanks again!)
Well, amazing and slightly frightening perspective at the same time. But as I. Tsamardinos pointed out, the role of human data scientist will shift focus to data preparation and outcome evaluation, as it should be. The image of sci-fi genuses fostering their artificial "child" (HAL or Mr. Data) comes to mind. For this to work instantly I see an obstacle in the computational power needed for ML processes. Furthermore, prediction is only part of the story, to be exaxt, the second part. Knowing what constitutes a model is implicit in any scientific field aiming at explaining the world and it gains importance where model based decisions have severe consequences for individuals' lives, e.g. bank lending or forensic risk assessment. But yeah, having AutoML would leave more time for other nice things in life, too. Looking forward to seeing how the story evolves.
Thanks for the video, It is really interesting to watch your video and I can easily understand the statistics I did not understand before, could you also make one for PLS and PLS-DA?
I have a dataset including dependent variable and multiple independent variables (features), Is there any platform I can use to make machine learning training with the most frequently used models like the random forest, SVM, naive Bayes ... and return a comparison table to show the performance for these models?
To add to what Josh just said, we have just added a new feature that reports, in addition to the detailed performance of the best performing model and the best interpretable model, a report of the best performance of each machine learning algorithm we test. Would this be what you want?
Sorry to hijack this comment section, but I was wondering whether StatQuest talks about the DBSCAN algorithm in one of his videos (since I tried searching for it, but couldn't find any). TY in advance.
Glad you liked it Moises - Ioannis did a detailed video on using JADBIO with Covid19 data that is also available as a hands-on tutorial, if you would like to try it. (ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-lHCjEmlOigc.html). You can also sign up to use JADBIO at our web site. It is a friendly trial subscription (no automatic access to your credit card)
I see this as the ode45 of machine learning - incredibly powerful but issues surrounding understanding of underlying system, instability etc. I don't think it's good to have people using tools they don't understand. Machine learning models for the most part are black boxes, interpretability techniques such as LIME/SHAP are not a substitute for model understanding - feature importance of an instance isn't an understanding. We already see this kind of sausage machine, handle cranking mentality in professionals in electronics with circuit simulation etc, the more trust we put in a system we don't ourselves understand, the more we use tools unknown to us - the more dangerous they are
To a certain extant, we address this at 10:46. Using a high level programming language, like Python, which takes care of memory management and other low level things that used to be programmed by hand in assembly language, doesn't make one a bad programmer. In fact, it makes you a programmer that is in demand, job wise. Likewise, learning high level ML tools doesn't make you bad at ML - but you still need to know what you're doing.
I remember when people used to say the same sort of things about bioinformatics tools. Gradually the tools and the users evolved in a way that domain experts outside of bioinformatics were able to use user-friendly software tools to enable their research. The bioinformaticians were then able to focus on much more interesting challenges. I think the same will also be true for AutoML, but I agree with you, prior to the evolution, it is possible for a novice to make erroneous conclusions, and it is the responsibility of the developers of friendly tools to provide safety guards against that naivety. #StatQuest, also helps a lot!
Do I need to watch or study the full playlist of machine learning or i have to learn automl only? thnks for replying.. your videos and explanation is awesome. thnk uh
@@pankajmodi8009 It really depends on what you want to get out of it. At a bare minimum, you should be familiar the general idea of ML and my four "Machine Learning Fundamentals". These are the first 5 videos in the ML section on this page: statquest.org/video-index/
I'm not sure. 90% of what we do is format/clean data and deal outliers etc. Only a small percentage of time is spent fitting models to the data. This makes that part easier.