[0:00] so you want to learn machine learning
[0:01] and you somehow ended up here well I've
[0:03] got good news and bad news the bad news
[0:06] I'm not here to sell you some become an
[0:08] ml engineer in 3 months fantasy could it
[0:10] happen sure people also win the lottery
[0:13] the good news I've been exactly where
[0:16] you are about 8 years ago I thought I
[0:18] knew enough Basics to just apply for
[0:20] jobs and learn the rest while working
[0:22] spoiler alert I failed miserably but
[0:25] those failures taught me exactly what
[0:26] works and what doesn't and that's what
[0:28] I'm sharing today I've since taught all
[0:30] of this to hundreds of students in
[0:31] several countries so I think I have a
[0:33] good idea on what works and what doesn't
[0:35] hopefully I can save you months of
[0:37] frustration by showing you the smart way
[0:39] to learn machine learning learning data
[0:41] science was one of the best decisions I
[0:43] ever made and most of you can do it and
[0:45] you will learn some cool stuff on the
[0:47] way that even if you don't become a data
[0:49] scientist or machine learning engineer
[0:50] you will have learned programming how to
[0:52] build apps how to analyze and visualize
[0:55] data you will have strong statistics and
[0:57] research skills and be able to
[0:58] communicate data clearly
[1:00] many amazing job options will be open to
[1:02] you even if you don't become a data
[1:03] scientist but you will have to work hard
[1:06] but what should you work hard on and how
[1:08] do we even start that's what I'm here to
[1:10] tell you learning how to learn before we
[1:13] even touch machine learning let's talk
[1:15] about something crucial learning how to
[1:17] learn why because machine learning and
[1:20] AI like most things in Tech are
[1:22] constantly evolving and what matters
[1:23] isn't just what you know but how quickly
[1:25] you can adapt and learn new things a
[1:27] little secret I actually suck at at
[1:30] programming and algorithms but I am
[1:31] really good at learning new stuff here's
[1:33] why this matters specifically for
[1:34] machine learning technology changes fast
[1:37] new platforms and Frameworks drop
[1:38] constantly and new papers come out daily
[1:41] what's hot today might be obsolete
[1:42] tomorrow problem solving is everything
[1:45] machine learning isn't about memorizing
[1:46] algorithms it's about understanding data
[1:48] and patterns about breaking down complex
[1:50] problems and finding Creative Solutions
[1:52] confidence AKA don't get overwhelmed or
[1:55] scared of big problems people who know
[1:57] how to learn and problem solve don't get
[1:58] paralyzed when faced with a new big
[2:00] problem they develop a strategy for how
[2:02] to look at a new problem and break it
[2:04] down into manageable problems they have
[2:05] faced before they know how to look up
[2:07] Solutions and find tools necessary to
[2:09] solve new problems they adapt more
[2:11] quickly when Tech changes efficiency if
[2:14] you know how to learn you won't waste
[2:16] time on unnecessary things time is money
[2:19] learn what you actually need to get
[2:21] where you want to be there's no one size
[2:23] fits all solution for learning something
[2:25] it depends on your style of learning but
[2:26] also on your goals not everyone needs to
[2:29] learn everything so how do you learn how
[2:31] to learn this one you kind of have to
[2:33] figure out for yourself because what
[2:35] works for one person doesn't necessarily
[2:37] work for the next some people learn well
[2:39] with graphs and diagrams others with
[2:41] text others maybe with voice notes some
[2:43] people need to understand the theory
[2:44] before applying it others need to jump
[2:46] right in and use an algorithm before
[2:48] asking what it actually does in this
[2:51] video I will try to show you what worked
[2:52] for me while giving you resources that I
[2:54] believe will get you there as quickly as
[2:55] possible I will just mention a principle
[2:57] that has helped me a lot throughout my
[2:59] career the the Paro principle sometimes
[3:01] called 8020 principle it says that 80%
[3:04] of the results come from 20% of the
[3:06] effort constantly ask yourself why am I
[3:09] doing this is this actually getting me
[3:11] where I want to be or can I do something
[3:13] more useful with my time well the answer
[3:15] to this question isn't always the same
[3:17] for everyone I will try to now give you
[3:19] the 20% of the work that would have
[3:20] gotten me 80% of the way to becoming a
[3:23] data scientist adapt as needed but where
[3:25] do I start now let's build your machine
[3:28] Learning Foundation the right way
[3:30] here's your road map python while the
[3:33] next skill is at least as important I
[3:34] would start with learning python the
[3:36] main reason is that you will get a
[3:37] feeling of achievement fairly quickly
[3:39] and python is super simple why python
[3:42] python is the main language of data
[3:44] analytics data science and machine
[3:45] learning while also being a full-fledged
[3:47] programming language allowing you to
[3:49] write scripts build apps and websites
[3:51] and much more python will allow you to
[3:53] actually start writing real code within
[3:55] days without having to learn super
[3:57] complicated computer science Concepts
[3:59] like pointers memory allocation and
[4:00] garbage collection also with python you
[4:03] will be able to get a job as a
[4:04] programmer or data analyst or web
[4:06] developer even if you don't learn all
[4:08] the hard machine learning stuff I
[4:10] suggest you first install jupyter
[4:12] notebooks as they make learning much
[4:13] easier and jupyter notebooks are also a
[4:15] core tool for data analysts and data
[4:17] scientists all over the world then learn
[4:19] about these Core Concepts programming
[4:21] fundamentals basic syntax indentation
[4:24] rules comments and so on variables math
[4:26] if else Loops printing data types like
[4:29] strings ins floats booleans lists
[4:31] dictionaries functions classes and
[4:33] objects modules packages and importing
[4:36] do a pandas tutorial pandas is Python's
[4:38] primary data manipulation Library built
[4:40] for handling tabular data through data
[4:42] frame objects imagine it as Excel
[4:44] spreadsheets on steroids it will be your
[4:46] main tool for data analysis cleaning and
[4:48] transformation with powerful functions
[4:50] for merging reshaping and analyzing data
[4:53] the library strength lies in combining
[4:56] the power of numpy arrays with
[4:57] spreadsheet like functionality and SQL
[5:00] database like joints it also comes with
[5:03] built-in plotting functionality built on
[5:04] top of Python's powerful met plot lib
[5:06] libraries pandas is a true data analysis
[5:09] Powerhouse if you truly Master pandas
[5:11] you will excel at most data analysis
[5:13] positions in the world also because
[5:15] exploratory data analysis and data
[5:17] preparation are about 60 to 80% of a
[5:19] data scientist job it will also lay the
[5:21] foundation for that 8020 principle
[5:24] remember your First Data analysis
[5:26] project so before you get into any
[5:28] machine learning I would take the time
[5:30] here to work on an actual project to
[5:31] deepen your python pandas and data
[5:33] analysis knowledge as I mentioned in my
[5:35] previous videos real projects beat
[5:37] tutorials at developing a good data
[5:39] scientist find some data you want to
[5:41] analyze maybe from one of your old jobs
[5:43] or school maybe you can export some data
[5:45] from your favorite health tracker or ask
[5:47] some friends if they have some data they
[5:49] want analyzed or maybe download public
[5:51] data from the government the World Bank
[5:53] or a nonprofit any topic you're
[5:55] interested in it could be economics
[5:58] Sports politics video games the board
[5:59] games this last one was a passion of one
[6:02] of my former students work on importing
[6:04] the data into pandas clean up the data
[6:07] make the units uniform decide what to do
[6:09] about missing data and outliers plot the
[6:11] different variables look at correlations
[6:13] between variables and come up with some
[6:15] hypothesis about the data and test them
[6:17] by making more plots turn your results
[6:19] into a slideshow with nice graphs that
[6:21] tell a story that you can present to
[6:22] friends and family Pro tip Jupiter
[6:24] notebooks with data and plots can be
[6:26] turned directly into a slideshow this
[6:28] will also be the first project for your
[6:29] port portfolio which you can show when
[6:30] applying for jobs as a data analyst
[6:33] essential math for machine learning this
[6:36] might be the part that most of you fear
[6:37] the most but I think it is the most
[6:38] important part for anyone wanting to
[6:40] learn machine learning you should take
[6:42] this seriously you don't need to be a
[6:44] math genius or know about all of math to
[6:46] become good at machine learning but you
[6:48] need to really understand the Core
[6:49] Concepts from the areas I'm about to
[6:51] mention for more details on math for
[6:53] machine learning check out my video on
[6:54] the topic basic statistics and
[6:57] probability this for me is the most
[6:59] important Branch as a data analyst and
[7:01] data scientist now there are many online
[7:03] resources for statistics but I highly
[7:05] suggest taking the Con Academy
[7:07] statistics and probability course this
[7:09] course is completely free and is the one
[7:11] I took when I prepared for my first job
[7:12] as a data scientist the full course is
[7:14] probably around 50 hours of content so
[7:16] if you have prior math knowledge you
[7:18] probably won't spend more than 100 hours
[7:19] on this but it might take you longer
[7:21] that's around 2 3 weeks of full-time
[7:23] self-study more if much of this is
[7:25] completely new to you but please take
[7:27] the time to do this it will make
[7:29] everything that follows so much easier
[7:30] and save you much more than 100 hours of
[7:32] headaches later on ideally while you
[7:35] learn new Concepts here you go to your
[7:37] data set from the previous data analysis
[7:39] project phase and try to apply them
[7:40] there to deepen your intuition linear
[7:42] algebra
[7:44] fundamentals while also important linear
[7:46] algebra for machine learning is much
[7:47] more about learning some tools and rules
[7:49] this should be much quicker than
[7:50] learning probability and statistics
[7:52] Concepts the main thing you want to
[7:54] learn is how to operate with vectors and
[7:56] matrices and learn what the different
[7:57] operations mean this is more about
[7:59] mathem iCal tools and notations than
[8:00] Concepts I think learning this will take
[8:03] about a quarter to a third of the time
[8:05] it took you to learn the statistics
[8:06] Concepts so one or two weeks of studying
[8:09] should be enough for people with prior
[8:10] math
[8:11] knowledge I will also leave the link to
[8:13] the Khan Academy linear algebra course
[8:15] in the description calculus here again
[8:18] it's about learning some tools but also
[8:20] understanding what derivatives are
[8:21] conceptually and how they help in
[8:22] optimization problems you should really
[8:24] understand how functions and their
[8:26] derivatives work and know the basic
[8:27] rules of differentiation like the chain
[8:29] rule I would again calculate with one or
[8:31] two weeks if you have prior math
[8:33] knowledge I will also leave a con
[8:34] Academy Link in the description Pro tip
[8:38] you need working knowledge not a math
[8:40] PhD focus on intuition over proofs spend
[8:44] most your time on statistical concepts
[8:46] for linear algebra and calculus focus on
[8:48] learning the tools like Matrix
[8:49] operations and how to take the
[8:51] derivative of a function the core
[8:53] machine learning Concepts and algorithms
[8:56] now here's where many people mess up
[8:58] they jump straight to Deep learning but
[9:00] that's a mistake in my opinion you
[9:02] should spend most of your time on simple
[9:03] algorithms the reasons for that are
[9:05] manifold and discussed in my previous
[9:07] videos but basically many problems don't
[9:09] require complicated solution simple
[9:12] algorithms like linear regression are
[9:13] quicker to run they're more
[9:15] generalizable more interpretable and
[9:17] easier to learn from and communicate and
[9:19] more importantly these algorithms form
[9:21] the basis for the more complicated
[9:23] algorithms like neural networks so truly
[9:25] understanding them will help you
[9:27] understand the more complicated
[9:28] algorithms better too check out my video
[9:30] on machine learning algorithms but
[9:32] basically before getting into neural
[9:34] networks or even svm make sure you
[9:36] understand how linear regression and
[9:37] logistic regression work then look at
[9:39] decision trees and Ensemble algorithms
[9:42] like random forests and gradient
[9:43] boosting I learned most of these topics
[9:45] from the book an introduction to
[9:47] statistical learning the majority of you
[9:49] will prefer learning from videos so just
[9:51] watch the Youtube video series about
[9:53] this book by the authors themselves
[9:55] completely for free on YouTube I will
[9:58] leave a link in the description
[9:59] all the videos together are about 20
[10:01] hours but since you will want to pause
[10:03] and take notes and read up on certain
[10:05] Concepts I think this will be another
[10:07] 100 hours or so of study time so that
[10:09] should be another two or more weeks of
[10:11] full-time self-study all these numbers
[10:13] are estimates as everyone learns at
[10:15] different speeds do a scikit learn
[10:17] tutorial scit learn is the number one
[10:19] machine learning library in the world
[10:21] for basic machine learning algorithms
[10:23] you can do a basic sklearn tutorial in a
[10:25] day or two and the good thing is that
[10:27] the simple and consistent syntax makes
[10:29] it such that once you know how to use
[10:31] the library for one algorithm you know
[10:33] how to use it for any algorithm as long
[10:35] as you know what algorithm is meant for
[10:36] what which you now know because you just
[10:38] learned it pyit learn also comes with
[10:41] great documentation and toy data sets to
[10:43] play around with I suggest you start
[10:44] using psyit learn while you are learning
[10:46] about the algorithms in the statistical
[10:48] learning course the genius move while
[10:51] you learn about the theory behind new
[10:52] algorithms for example going through the
[10:54] statistical learning course and starting
[10:56] with linear regression Implement and use
[10:58] the algorithm in the following three
[10:59] three ways implemented from scratch
[11:01] using basic python implemented using
[11:04] scikit learn using a toy data
[11:07] set then use both your own
[11:09] implementation and the syit learn
[11:10] toolkit to try out the algorithm on a
[11:12] real data set that you have prepared
[11:13] yourself now there's a common Pitfall
[11:15] that many beginners get stuck in
[11:17] tutorial hell it's where you essentially
[11:20] just keep following tutorials without
[11:21] striking out on your own and actually
[11:23] building something most learning comes
[11:25] from the trial and error of building an
[11:26] application so if you always follow a
[11:28] tutorial you get Stu at a basic level
[11:31] how do you not get stuck there do just
[11:32] one or two tutorials per area Max and
[11:35] then work on a real project your first
[11:37] machine learning project here you can
[11:39] either continue with your data analysis
[11:41] project from before or find a new more
[11:43] interesting data set for a machine
[11:44] learning project but don't forget to
[11:46] still use pandas to do an exploratory
[11:48] data analysis to prepare your data for
[11:50] modeling and form hypothesis about your
[11:52] data more often than not the goal of a
[11:55] machine learning project will be to
[11:56] predict some variable from other
[11:57] variables research the industry of your
[11:59] project a bit look at the data and make
[12:02] some hypothesis about what might
[12:03] influence your target variable either
[12:05] based on intuition about the industry or
[12:07] from looking at correlations and Scatter
[12:10] Plots of different
[12:11] variables design new features based on
[12:13] your knowledge of the problem then start
[12:15] modeling but start with simple
[12:16] algorithms like linear regression
[12:18] logistic regression and decision trees
[12:21] then move on to more complex algorithms
[12:22] like svm random forests or gradient
[12:25] boosting note how as complexity
[12:26] increases accuracy usually increases but
[12:29] interpretability decreases your goal is
[12:31] usually to find a sweet spot also don't
[12:33] forget over fitting as you increase in
[12:35] model complexity keep validation and
[12:37] test sets aside before starting to model
[12:39] and compare your models using the test
[12:41] set at the very end many times the more
[12:44] complex algorithms don't look as good
[12:45] anymore once you use the final test set
[12:48] it might be a good idea to work with
[12:50] data sets that have been published on
[12:51] sites like kaggle to then compare your
[12:53] Solutions and accuracies to other
[12:54] people's Solutions and get an idea of
[12:57] how well your models are in comparison
[12:58] to others but don't get frustrated a lot
[13:01] of people on kaggle are professionals
[13:03] with years of experience if you get
[13:05] anywhere close in accuracy you should be
[13:07] happy don't know what to work on you can
[13:09] start with a tutorial but instead of
[13:11] following it directly after building the
[13:12] core features add some features change
[13:15] some features swap out the data set and
[13:17] try to break your code and then fix it
[13:19] this is one of the best ways to learn
[13:21] while not getting stuck in tutorial hell
[13:24] collaborate and share your projects with
[13:25] others learning ML and isolation is the
[13:28] slowest way to learn instead find coding
[13:30] buddies to work on a project with
[13:32] present your work to friends and family
[13:33] or post it publicly on GitHub or in
[13:35] machine learning communities have
[13:37] someone more advanced than you give you
[13:38] feedback this will speed up your
[13:40] learning 10 times don't know anyone to
[13:42] work on a project with participate in a
[13:44] hackathon or write to people with
[13:46] similar interests on kaggle GitHub
[13:48] Discord Reddit LinkedIn Etc the
[13:50] connections you form this way will not
[13:52] only help you learn better but boost
[13:53] your career in unexpected ways check out
[13:55] my most recent video to learn more about
[13:57] the importance of networking and data
[13:59] science Advanced topics only now should
[14:01] you look at more advanced topics deep
[14:03] learning architectures cnns for computer
[14:06] vision RNN for sequential data or
[14:08] Transformers for NLP Advanced
[14:10] optimization techniques model deployment
[14:12] strategies and the latest research
[14:14] papers remember learn these by need not
[14:17] by fomo you don't need to know
[14:19] everything just learn these techniques
[14:20] if they are important to your project
[14:22] here some dos and don'ts don't don't get
[14:25] stuck in tutorial hell don't try to
[14:27] memorize everything don't learn in
[14:29] isolation don't chase every new trend
[14:32] don't copypaste code without
[14:33] understanding don't try to learn every
[14:35] new fancy tool or research paper instead
[14:38] build real projects focus on
[14:40] understanding share your progress join
[14:42] communities Master fundamentals first
[14:46] Implement from scratch learn by doing if
[14:49] you found this video helpful share it
[14:50] with someone who you think might also
[14:52] like it and get started on one of the
[14:54] tutorials in the description or on this
[14:56] very Channel also consider liking the
[14:58] video and subscri subcribing to be
[14:59] notified about similar content in the
[15:01] future thanks for watching