Data scientists should be more end-to-end | Machine Learning Monthly September 2020

Every month I review the latest and greatest resources in the world of machine learning in a newsletter called Machine Learning Monthly. This is the video version of that newsletter for September 2020.
And as usual, it's been a massive month on tour.
Read the blog post - zerotomastery.io/blog/machine...
Learn ML (beginner-friendly course) - dbourke.link/mlcourse
Updates on my work - dbourke.link/newsletter
Timestamps:
0:00 - Intro and hello
0:57 - How I'd start learning ML again video
1:47 - Shivan's project to analyze my KZbin data
4:16 - Free advanced NLP with spaCy course
5:43 - Explosion AI (one of my favourite AI companies)
6:15 - Text classification with TFIDF vs Word2Vec vs BERT
7:16 - TFIDF formula
8:35 - Super efficient on-device NLP
11:10 - Story of working on a NLP ML project
13:07 - Machine Learning from Scratch book by Danny Friedman
15:33 - 42 day project framework
16:05 - Learning Math for Machine Learning
17:35 - One of my favourite quotes on learning
19:06 - Four great questions for any ML project
19:48 - Data scientists should be more end-to-end
22:43 - Story of being held back on a project
25:34 - GitHub Actions for MLOps & Data Science
27:15 - Microsoft DeepSpeed framework for faster and more efficient model training
30:32 - Roboflow's Ultimate Guide to Object Detection in 2020
33:37 - Using AI and computer vision for agriculture
36:25 - My favourite resource for the month
37:00 - Outro
Support on Patreon - bit.ly/mrdbourkepatreon
Connect elsewhere:
Web - dbourke.link/web
Quora - dbourke.link/quora
Medium - dbourke.link/medium
Twitter - dbourke.link/twitter
LinkedIn - dbourke.link/linkedin

[@mrdbourke]

my bad. this one went a bit long... I just get so excited talking about machine learning and telling stories from projects I've worked on. will make sure the next one is

[@aniketpatil2470]

Duration doesn't affect much, when the content is worth watching

[@shaheenahmed3506]

I personally thought it was a really good length, I put this on in the background while I'm relaxing, so could have listened for longer! Question though Daniel, how on earth do you get time to do all of these extra activities alongside your full-time job?

[@egehurturk6209]

Since the content is superb, time doesn’t matter for me

[@ashikshafi1765]

It was great tbh I enjoyed watching this soo much info and + storytelling got us engaged🔥 Thanks for another masterpiece brother your killing it ❤️

[@ajmalali1875]

For me, to be honest, I prefer short, digestible, < 20 mins. videos.

[@ShivanS]

Hey Daniel, once again great video and thank you for the mention! Keep this up.

[@mrdbourke]

Hey Shivan! Glad you enjoyed legend. Thank you for sharing your work with me

[@agustinsantellanfernandez5587]

This videos are really inspirational. Positve actitud while starting any project in life. Keep doing this videos!

[@mrdbourke]

Thank you Agustin! Plenty more videos to come :)

[@preetimurmu851]

Very nice video, i really enjoyed it and surely try to get much from these videos. Thank you for making it and sharing with the world.

[@prod.kashkari3075]

Ugh I wish I could just have the summer again where I could continue my learning school takes up so much time 😤

[@mrdbourke]

Anytime is a good time for learning

[@prod.kashkari3075]

@@mrdbourke true. Just feel like tech moves so fast that by the time I get to learning something it may be outdated/new stuff came out which is better. Don’t think this will be the c are for tensorflow and pytorch for a good period of time but idk.

[@dibyajyotiparida7316]

Thanks this will be very helpful😃

[@user-ph4cw7ym2j]

i respect you bro you are so talented

[@d3v487]

Awesome Guide Daniel As always... But we want you to upload some hands-on Transformer implementation (BERT, GPT...etc) so that it'll very helpful for others. I have enrolled in your course and its outstanding :) So please make some videos regarding transformers. Love from India.❤

[@nhesneilragasa1953]

Keep doing this monthly!

[@mrdbourke]

Sure thing!

[@daniel_romero]

Thanks Daniel, fantastic content! =)

[@mrdbourke]

Glad you enjoyed Daniel! PS great name

[@JSmoove90]

Machine Learning is the key I actually put more effort to it then any other thing out there because its so broad and I learn something new about it everyday...Thanks good vid

[@mrdbourke]

You and me both and thank you for the kind words

[@sajanah1253]

YOOO i was just watching your airbnb series. When do we get to see your new machine learning PC at work???? Btw im going to start your ML course at ZTM! (just finishing up the python course atm)

[@mrdbourke]

I’m working on a couple of projects right now but when they’re done, I’ll make another series using the ML PC!

[@muhammadsifaulrizky2298]

Nice video as always, really needs for my jobs who currently on NLP topics, my main issues are for cleansing, because my language, Bahasa Indonesia, is so many words, and need for cleansing. Thank you for your videos.

[@mrdbourke]

Thank you Muhammad! Glad you enjoyed :)

[@tech4028]

Can you make a video on how to deploy *pretrained* models like T5 (huggingface + pytorch), please?

[@vikramc.n.1121]

Hey sir, looking cool in this shirt, thank you so much for the video by the way

[@johnwt7333]

He looks better without

[@mrdbourke]

Hey Vikram! Thank you for the kind words legend

[@vikramc.n.1121]

​@@mrdbourke You are the real legend sir, thank you for being such a motivation to everyone.

[@cakraocha]

This is a really good session (at least that's what I called it). Thanks Daniel! I always confused about whether how far should I study to be a generalist data scientist. Definitely will check the article you've shared. Thanks a lot and keep it up!

[@mrdbourke]

Glad you enjoyed Gede! Enjoy the article my friend

[@muhammadsifaulrizky2298]

Woah you here too, nice to see you on the comments, Daniel is one of my favorite channels for getting resource about Data Science, especially now focusing on Machine Learning.

[@cakraocha]

@@muhammadsifaulrizky2298 long time no see mate!

[@waeldimassi3355]

Really cool video !!

[@mrdbourke]

Thank you Wael!

[@kirtipandya4618]

Hi Daniel, thank you very much for wonderful content. Which framework you personally like more? Fastai or TF?

[@mrdbourke]

Thank you Kirti! I love them both. I’m deep in TensorFlow right now but will probably get more familiar with fastai V2 over the Summer

[@TJ-ms7rn]

Awesome video!

[@mrdbourke]

Thank you Tyler!

[@bigbrave2440]

Is there any projects you are working on? Just out of curiosity. How do you stumble to these info and resources? Is there a link you have with resources that other ppl can put resources, if there is what is it as i have collected couple myself and i would like to contribute.

[@mrdbourke]

I’m working on a couple of projects I’m not ready to share yet, but they’ll be released soon! I come across these during my browsing/researching machine learning online You can send a link to me directly if you want! [e-mail] Otherwise www.madewithml.com is a great place to share things

[@user-to7dp2uc3o]

Great Video!

[@mrdbourke]

Thank you!

[@vyomeshojha6704]

Hey Daniel I have been watching and reading your content and have been in ML for about 5 months now. But now I feel like I am now stuck and don't know where to go as I have started with 3 -4 books and a DL course on Coursera but couldn't finish it, I also looked for projects to do some Hands-on but couldn't find any luck and now I have been kind of lost. From where to restart learning that's what I am not able to figure out. Suggestions! Thanks!

[@tyrantula767]

I would highly recommend dataquest.io if you want to learn python, numpy & pandas, sql, web scraping, data cleaning and visualization, and the basics of ML and DL. If you already know python and many of it’s data science packages, I’d suggest the book “Introduction to Machine Learning with Python” by O’Reilly Media for the coding part of ML & and “An Introduction to Statistical Learning” for the maths. If ISL is too hard, dataquest.io does a great job at explaining the maths a little bit simpler. Find some use cases for each ML model, and then try to remake the model using new data.

[@vyomeshojha6704]

@@tyrantula767 thanks for your suggestion Phil.. :-)

[@sunayanak5438]

Hi awesome video I have a small question. I have completed basic python (list,tuple,dict,functions,etc) do I need to know about other things as well in my machine learning journey with respect to python? Like decorators n oops concepts as well?? Please do reply

[@mrdbourke]

Thank you Sunayana. Yes, you will eventually use those concepts, learn them as you need them. Start by writing as much machine learning code as possible.

[@deepakkumarpatel5728]

hey Daniel i have been studing data science for last 6 months but apart from studying I haven't done anything , I haven't written and blog I haven't uploaded any thing in github etc, I am afraid of writing any thing I am afraid of almost everything, I am afraid of trying thing what should i do i have no idea, i forgot what i have learned so far as well

[@abhishekprajapat415]

Hi Bro, although I am not DENIAL still I would like to help. There are a couple of things I would tell u to do:- NOTE:- HOPE U KNOW PYTHON #################################### 1) It's not strictly related to data science but always uploads your code to GitHub when u r learning and also learn git and after spending that week on this download your first cheatsheet of Git. #################################### 2) Now whatever you learn write the code about it. Writing code is a must. The only theory will make u lose interest and the knowledge will be less stable. #################################### 3) After each week (on a weekday) write a word document about what u learned and what was good/different in it. basically create your own cheatsheet of the week. #################################### 4) After you have completed the basics of Machine learning - ( Pandas, NumPy, matplotlib, basic machine learning algorithms, and a very basic neural network ). for this, you could do the first 3 courses of Coursera Specialization "Applied Data Science with Python" #################################### 5) At this moment what you know is the very basics of this field. Now I want you to just explore a little bit of neural networks for supervised learning only. Do the supervised part from the "Deep learning A-Z" course on Udemy. #################################### 6) At this moment you should feel an interest either towards IMAGE DATA or TEXT DATA. Based on that choose a fiend either Computer-Vision or NLP. After choosing that pursue that field from the very basics. #################################### HOPE THIS HELPS !!!

[@mrdbourke]

Some great advice from Abhishek! I agree with all of the points. And Deepak, when it comes to creating and doing important work, find somewhere to put your fear, if it doesn’t help you, it’s not needed. You might like this article: www.mrdbourke.com/you-dont-need-permission/

[@MasterCivv]

Favorite optimizer is LAMB :D

[@mrdbourke]

Ooo, I haven’t even heard of that one... I’ll have to look it up

[@MasterCivv]

@@mrdbourke it is the one used to make Bert converge at really large batch size! Also the zero optimiser trick is done offloading optimiser States to the CPU, this frees up 5+ times the memory on the gpu allowing larger models! I am really looking forward to the next monthly, it is really interesting and useful to keep up with nice work done in the ml world!

[@akhandpratap__]

4th viewer 😇😇😇😇 ( Likes more than views, bro u r killing it )

[@mrdbourke]

Thank you Akhand!

[@semrana1986]

yup agreed in the applied world I rather see it as Data Science and Engineering (Soft Eng + MLE)

[@mrdbourke]

Too right! ML Engineering is 90% engineering & 10% ML

[@demetrio6159]

Is it possible to work as a researcher in machine learning remotely?

[@mrdbourke]

Yes, of course, you have the internet, you can start researching now. If you want to be hired, I’d relentlessly share your research through your own website/GitHub and use it to demonstrate to companies/academies what you can do.

[@demetrio6159]

@@mrdbourke Thanks man! I doing my website for shared my knowledge. I'm hope to get a job in the field soon.

[@mholmeslinde]

i really like your videos, and i always appreciate your perspective on all matters machine learning. BUT - your standing desk is too high! you're young and fit, so you might not feel it for a while, but having to lift your arms to that height to use the keyboard/mouse is rough on the body. try lowering your desk to the point where you can have your shoulders relaxed, your elbows at your sides, and your arms at a 90 angle. You can honestly get away with a little bit lower than that, even. Then, just raise your monitors a bit higher to get them back up to eye level. I know this I am an old man (mid 30s) with lots of RSI risk from being a professional instrumentalist, and I almost blew up my right rotator cuff with my standing desk too high.

[@mrdbourke]

Thank you for the advice Matthew! I’ll try out the lower position!

[@mholmeslinde]

@@mrdbourke excellent. i hope it works for you. as a fitness nerd, I'm sure you know that prehab > rehab.

[@qwertyasdfg6403]

how much time does it takes to learn machine learning from scratch ???????

[@Shubham-jr8nc]

infinite!

[@abhishekprajapat415]

Not that much. but when u start the field I would recommend you go more Hands-on at first and don't touch the theory. after 2 months when u have done a little bit of ML start from foundations.

[@qwertyasdfg6403]

@@abhishekprajapat415 thank you

[@Matt-yp7io]

1 day according to this guy

[@mrdbourke]

As long as you need. But if you’re dedicated you can become pretty skilful in 6-12 months. However when it comes to ML, the learning journey never stops.

[@mehdi179]

Hi. I have an idea. What about doing videos about contributing on github and when some of us get stuck, he will comment the problem? For example, I am stuck today trying to complete this course on github. So I let it go. github.com/moatazeldebsy/TestCafe-TAU

[@saadjadoon6397]

Hello

[@mrdbourke]

Hello hello

[@abhi6841]

First

[@mrdbourke]

You’re quick!

[@skinnyboystudios9722]

Most of what you speak about is pure AI. I don't see how its related to data science

[@mrdbourke]

What’s the difference? I see data science as part of AI and vice versa

[@skinnyboystudios9722]

@@mrdbourke AI is part data science but not the reverse. Remember AI is a subset of computer science. What you study in Computer science are three main things, Computation(the procedure of algorithms), Computability(proof of validity of algorithms and 3 complexity(measure of algorithms. Even if you had to study data science by evaluating it from AI techniques, you still have to follow those three principles of computer science and at the end it has nothing to do with data but analysis of that algorithm... Problem is that most people especially youtubers fail to explain to beginners the difference between DS and AI and end up confusing new students. Even videos that try to speak about AI, end up talking about Deep Learning and not explaining the entire AI filed. There was even a guy who visited me claiming to be an AI expert but did not know about classical methods like heuristics, constraint satisfaction problems, hill climbing, simulated annealing, Knowledge based systems and so on. He only thought all AI was machine learning before I showed him that in game development we do not really need machine learning as classical AI works very well... That just being said, I really enjoy your videos and hope that you do a video on AI and DS vs AI breaking everything step by step not only talking about deep learning as if other types of AI are not as important. Thank you and sorry for the speech

[@Scotttarlow]

@@mrdbourke the biggest difference is that most data scientists do not work in machine learning. They are focused on measuring impact - causual modeling, design of experiments, and continuous improvement of technical products. The real goal of a data scientist is to provide business impact from data. Machine Learning and AI are as others stated small parts of data science as its currently practiced. Most data scientists (and highest paid data scientists) work on KPI development and other analytics task. This can easily be observed from the job market - at FAANG companies 80% of jobs are design of experiment and probabilistic modeling based jobs. Not to be an asshole, but your opinions sound like someone who actually doesn't work in the field .