シャドーイング練習: How to learn Machine Learning like a GENIUS and not waste time - 動画で英語スピーキングを学ぶ

Most people who try to learn machine learning quit.

Not because it's too hard, but because they waste months on the wrong things.

They binge watch lecture series, memorize math that they'll never use, and never actually build anything.

I've seen it hundreds of times, and in this video, I'm going to give you the exact learning path that actually works step by step, what to learn, what to skip,

and how to learn it in a way where you're not

spinning your wheels for six months with nothing to show for it.

And it's a shame when that happens because ML engineers are some of the highest paid people in tech right now.

And we're talking, you know, 150 to 200K plus just for starting roles.

So with that said, let's dive in

and let me discuss how to learn machine learning like a genius so that you don't give up.

Now, I want to start by discussing the trap that almost everybody falls into when they're trying to get into machine learning.

Now, that's trying to learn everything before they build anything.

Now people will literally spend three months learning linear algebra proofs and never train an ML model.

Now the fix here is to learn just enough theory to understand what's happening, but then immediately start applying this and actually building projects.

You learn way more when you're doing hands-on coding, you're solving problems, you're dealing with challenges, and then what you can always do is when you don't have enough theory, you can go back and learn it.

Now I made this mistake myself, I spent a ton of time on math learning all of the proofs, and I could have been six months further ahead had I just started building at the beginning.

So what you're going to see in this video is that a lot of this comes back to just build, and when you don't know what you need in order to build the thing that you're trying to build, you go back and you learn that theory.

So just keep that in mind, it's way better to fail building something first, we can always learn the theory later.

So with that in mind, let me go through step by step what you actually need to learn and what you can skip.

Now let's start with section one, which is Python fundamentals.

Now, this is absolutely essential.

You need to know Python.

Not Julia, not R, not another programming language.

Python is the first thing that you should learn even before you get into any math or theory.

Now, you don't need to be an expert here, but you do need to be comfortable with the following.

So variables, loops, functions, data structures, so things like lists, dictionary sets, file handling, basic object-oriented programming, and that's really it.

Of course, there's some other features that you're going to learn, but again, you can pick those up when you need to know them.

Don't spend months trying to learn all of it right here.

Realistically, you want to spend maybe three to four weeks getting comfortable with Python, and your goal should be able to write small programs on your own.

Read in a file, print out the outcome, right, or add something to it.

Create a simple CLI-based game.

You don't need to be a Python expert, but you do need to be comfortable with the basic syntax

so that you can read machine learning code and start writing basic scripts.

And that's something to keep in mind is that machine learning typically doesn't involve a massive amount of code.

It's usually smaller scripts and more having an understanding of what it is that you need to do.

So the faster you get into actually writing and building things, the more you're going to learn.

And to be honest, you'll pick up a lot of the fundamentals along the way.

Now with that in mind, there are a few key Python libraries that you are going to want to look at though, and that kind of tie into these fundamentals.

Now the first is going to be NumPy.

This is for arrays and math, and it's going to be used behind the scenes from a lot of other modules that you'll look at.

The next is pandas.

This is for data manipulation and looking at large amounts of data.

And then last is map plot lib.

This is for visualization, plots, graphs, etc. Now as a machine learning engineer, you're going to be using these almost every single day.

So understand the basics, know how to set them up, how to import them, what a data frame is, how to create a basic plot.

It really doesn't take that long.

And even just an hour tutorial is going to give you

a really solid base before you move on to the next stage.

So now let's talk about everybody's favorite subject, which is math, but specifically the math that you actually need to know.

Now, yes, there is a little bit of math that you are going to need for machine learning, but it's not as theoretical as a lot of people like to make it out to be.

And to be honest, this math is not super complex

and you don't need to be able to derive it or know all of the proofs.

You just need to have a high level understanding.

So here are the areas that actually matter.

And again, it might sound complicated, but I promise you just, you know, a few weeks of looking at this

and you're going to be much more comfortable with it than probably you would have imagined.

Okay.

So first we have a linear algebra basics.

So we're talking about what is a vector, what is a matrix, things like dot products.

Then we're looking at probability and statistics.

So what is a distribution, base theorem, mean, variance, okay?

Next, we look at calculus.

So just the basics, like what is a derivative, what is an integral, for example, and what is the concept of gradients and how optimization works.

And that's really it.

You don't need to be able to derive all of these algorithms or write them from scratch.

You just need to have looked at them before, understand at a fundamental level, kind of why they work or what they are, and be able to say, oh, derivative.

Yeah, I know what that is, could I drive this function?

Probably not, but I'm at least familiar with the concept.

If you know that, you're already going to be at a decent level for mathematics, and it's really only if you're going into super deep research

or really advanced machine learning that you're going to need to actually be good and solid in these topics.

I myself learned all of this in university probably five or six years ago now.

I forget almost all of it, but I at least am familiar with the word, and that already gives me enough basis to jump into some machine learning algorithms.

And as always here, if you find that you do need to know this math later, then that's fine.

Go and learn it.

Just don't make it a strong prerequisite that stops you from building immediately.

So now let's move on to the next section, which is core machine learning algorithms.

Now, once you've got the Python out of the way

and you know a little bit of those fundamentals when it comes to the math, and you can even learn those alongside these, you want to start looking at core machine learning.

Now, there's kind of some different categories that we look at here.

The first is supervised learning, then the second is unsupervised learning.

Now, within supervised learning, you have a few algorithms that you're going to want to be really comfortable with, and you're actually going to want to write, run, and train yourself.

Now, we're talking about linear regression, logistic regression, decision trees, random forests, SVM or support vector machines, and then K nearest neighbors.

Now, these are core machine learning algorithms.

They're very simple, and to be honest, most machine learning that you see nowadays uses some form or variant of these.

Now, after that, we have unsupervised learning.

Now, that's where we talk about things like K-means clustering and PCA, like dimensionality reduction.

There's also a few others here, but those are the most popular.

Now, for each of these algorithms, you want to understand what problem it solves, when to use it versus the alternatives that exist, and how to evaluate it.

So what's the accuracy?

What's the precision?

What's the recall?

How do you do cross-validation?

If you can understand that, then that puts you in a really solid place.

Now, in terms of playing with and using these algorithms, you're definitely going to want to be looking at scikit-learn.

This has a really clean API, great documentation, and it's really the go-to place for these kind of classical machine learning algorithms.

Now again, the key skill at this point is

that you want to know how to pick the right model for the right problem

and not just how to randomly run code.

Now in order to do this, like I said, use scikit-learn and then build some small projects, right?

100

So predict housing prices, classify emails, you know, cluster customer segments, look at real data sets and real problems.

101

There's all kinds of great tutorials out there.

102

And again, you just want to be super comfortable with the basics before we move on to anything more advanced.

103

Now, if at this point, you're convinced that you do want to learn machine learning, but you want to do it without wasting months stitching together random tutorials, then I really recommend Datacamp.

104

Now, I've partnered with them for this video, and that's because I've used Datacamp for years to level up my own Python and machine learning skills.

105

And what I like most about it is how hands-on it is.

106

Now, you're not just watching lectures, you're learning by doing.

107

So you're writing code in your browser, getting instant feedback and building real projects as you go.

108

If you want a complete curriculum that's focused on the model development side, the best place to start is their machine learning scientist track.

109

That gives you a structured path through supervised and unsupervised learning, feature engineering, model validation, XGBoost, NLP, and even deep learning with PyTorch.

110

So you're learning the skills that actually matter for real ML work.

111

And once you can build models, the next bottleneck is gonna be production.

112

Now that's where their machine learning engineer track comes in because it covers the stuff that most self-taught people struggle with.

113

So MLflow, Docker, data versioning, monitoring, drift, you know, CICD.

114

So you can go from my model works in a notebook to my model actually works in production.

115

Now you can also prove your skills with Datacamp's track credentials and certifications, which is great for your resume and for your LinkedIn.

116

Now Datacamp is trusted by over 19 million learners.

117

You can try both of these tracks for free with the link in the description.

118

And I genuinely wish that I had this kind of structure when I was starting it.

119

Anyways, it would have saved me a lot of time but now let's move on to step number four, which is deep learning.

120

So once you're comfortable with classical machine learning, you want to add on neural networks.

121

Now you want to start with the following concepts, what a neuron is, things like layers, activation functions, forward and backward passes,

122

loss functions, optimizers, and back propagation.

123

Now the framework that I suggest you look at here is PyTorch.

124

This is a lot more modern in 2026.

125

TensorFlow still exists, but PyTorch is really dominating research

126

and it's increasingly just more popular and kind of the production standards so you're probably going to be better off with that.

127

Now this framework allows you to do a ton but specifically to build out neural networks.

128

Now the key architectures that you want to look at

129

when it comes to neural networks is going to be feed forward neural networks, CNNs or convolutional neural networks for things like images, RNNs or recurrent neural networks,

130

and then LSTMs for sequences and then you want to look at things like transformers

131

and this is the current architecture behind every LLM, so it's probably going to be an interesting one to check out.

132

Now, you don't need to be able to build a transformer architecture from scratch on day one, but to understand things like attention

133

and how it works is going to make you a lot more effective even just with modern AI tools

134

and give you a non-black box understanding of what's actually going on.

135

Now, in terms of some projects that you can build here to help learn this, you can look at an image classifier with CNNs, sentiment analysis with a simple RNN,

136

and then you can even fine-tune a pre-trained model from something like hugging face.

137

Okay, so that's step four, neural networks.

138

Now let's move on to step five, which is the skills that actually get you hired.

139

Now this is where most roadmaps actually stop, but I want to go over things that you need to learn if you actually want to get a job, because all this is fine.

140

You're going to have fun.

141

You're going to build a build out models.

142

You know, it'll be really interesting, but if you go into a job interview, they're going to ask you about all of these things.

143

And if you don't know them or you don't have experience, you're going to be cut immediately.

144

So first we're looking at MLOps and specifically deployment.

145

So things like Docker, model serving, so fast API, Flask, inference servers, things like monitoring, CICD, and setting up basic ML pipelines.

146

As you get more advanced with machine learning, you actually realize that a lot of a machine learning engineer's role is not just to build the model,

147

but actually to deploy it so people can use it and it works in production.

148

Next, working with real data.

149

So data cleaning is literally 80% of machine learning work, and nobody warns you about this before.

150

When you're doing tutorials, you get all this clean, beautiful data with everything existing, but in the real world, you have messy data, missing values,

151

weird distributions, etc. So get comfortable with that early cleaning, parsing data.

152

Next, feature engineering.

153

Now often the difference between a mediocre model and a great one is feature engineering.

154

So having domain knowledge can actually matter more than the algorithm choice in many real world problems.

155

Next, version control, but specifically for ML.

156

So of course, we're talking about Git and GitHub, but also tools like ML flow and then weights and biases for experiment tracking.

157

Next, we talk about cloud platforms.

158

So you want to know at least one of AWS, GCP or Azure.

159

Personally, I recommend AWS, but really you can go with anything that you want here.

160

And then you're gonna be looking at tools like SageMaker, Vertex AI, Azure ML, you get the idea.

161

Now, overall, the people that get hired to do machine learning in the real world, right?

162

Who aren't just, you know, maybe junior interns who are just starting out, are hired because not only can they make the machine learning models, but they can serve them and use them in production.

163

That's oftentimes where the bottleneck is.

164

And especially now where a lot of the hard work is already done

165

and you're kind of just fine training or using existing models, that production side is super, super important.

166

So don't skip these skills.

167

So now let's talk about how to actually learn these skills effectively.

168

Now, the rule that I usually present is the 70-30 rule, where you want to spend 70% of your time on building projects and then 30% of your time on theory and courses.

169

Now, most people do the reverse of this and the ratios you can obviously play with, but generally speaking, you want to have at least double the amount of time actually building,

170

being hands-on, working on the computer, making mistakes, messing up, right?

171

Compared to when you're just watching tutorials or reading through theory.

172

Now, another thing that I highly suggest, especially in this field, is to learn in public.

173

It's not mandatory, but I do suggest that you post your projects on GitHub, write about what you're learning on LinkedIn, right?

174

And that's because specifically in ML, recruiters are hiring based on what you've built, not just what certifications that you have.

175

And this is a field where it does actually make sense to post a cool video about a neural network

176

that you trained or a cool project that you worked on

177

and it's not going to come across as I don't know kind of snobby

178

or arrogant it's something that's genuinely interesting

179

and I have a lot of friends in this space who are constantly talking about what they're doing

180

because it keeps them up to date keeps them relevant

181

and gives them a lot of opportunities again this is a

182

field where you need to stay adaptive you need to keep learning

183

if you keep posting and talking about it it at least shows

184

that you're in the industry and that you're active

185

and you're not just a dead LinkedIn profile that hasn't touched this in multiple years

186

next don't tutorial hop right pick one structured resource

187

and finish it before you jump to the next thing half completed courses really don't teach you anything

188

and you want to get that dopamine hit of actually completing something in full even

189

if it's not 100 the best resource next when you get stuck

190

and you absolutely will hear that's a signal to go

191

and learn some of the theory

192

so you have the context now you also have the end motivation for why you want to learn something

193

So now you can go and pick up the math.

194

Now you can go and learn another module.

195

Now you can go and pick up those pieces of syntax.

196

It's always easier to learn when you know why you're going to learn

197

that thing and you're not just learning it in isolation detached from a goal or from a project.

198

Next, make sure that you build end-to-end projects.

199

So data collection, cleaning, training, evaluation, deployment.

200

Doing this is going to teach you a lot more than doing 10 random Kaggle notebooks, okay?

201

Now in terms of a time estimate here, if you're super disciplined and you already have some programming knowledge, you can absolutely be job ready here in six to nine months.

202

It's not going to be six weeks, right?

203

But it's also not going to be three years, six to nine months of focused work.

204

And while it may seem like a lot in any field, that's really a short amount of time.

205

If you just spend it focusing on the right tasks and again, building before learning theory.

206

Now, my honest take here is that ML is one of the most rewarding skills that you can learn right now, but only if you learn it the right way.

207

The biggest trap is just staying in that learning mode and never building anything.

208

And that's because you don't need a PhD.

209

You don't need to understand every single paper, but you do need to actually be able to solve problems, which is the hard part and what a lot of people don't focus on.

210

Same thing like I talked about before, the deployment is a really important stage here.

211

So don't skip that or wait until you get into your first job interview.

212

If you follow the roadmap here, I guarantee that you're going to have some success.

213

Keep chipping away at it.

214

It's absolutely worth it.

215

With that said, guys, I hope you enjoyed.

216

Leave a like if you did.

217

Subscribe to the channel.

218

I will see you in the next video.

📺 同じチャンネル

✨ おすすめ動画

The Art of Learning Like a Genius (even if you're not one)

307

34 文

The Art of Learning Like a Genius (even if you're not one)

Business EnglishUniversity StudentsWorking Adults

265 文

How I Would Start Cybersecurity in 2026 (If Starting From Zero)

Business EnglishInterview EnglishUniversity Students

136 文

Is ML Engineering Dead? (AI vs ML Engineer)

Business EnglishInterview EnglishUniversity Students

12 文

How I’d Learn Coding in 2026 (Without Wasting Years)

Business EnglishUniversity StudentsWorking Adults

252 文

Memory is crazy expensive so I made my own…

Business EnglishUniversity StudentsWorking Adults

208 文

8 steps to get your sh** together

Business EnglishUniversity StudentsWorking Adults

260 文

It’s Actually Pretty Easy to Get Ahead of 99% of People with AI

Business EnglishUniversity StudentsWorking Adults

189 文

This Turns Your Expertise Into a YEAR of Binge-able YouTube Videos

Business EnglishUniversity StudentsWorking Adults

764 文

11. Introduction to Machine Learning

Business EnglishUniversity StudentsWorking Adults

90 文

If you’re an aspiring product designer, watch this (roadmap included)

Business EnglishInterview EnglishUniversity Students

354 文

How To Learn Any New Skill So Fast It’s Unfair

Business EnglishUniversity StudentsWorking Adults

196 文

I cloned 3 apps and now make $35K/month

Business EnglishInterview EnglishUniversity Students

この動画がスピーキング練習に最適な理由

この動画は、自然な英語のリズムと口語表現が豊富で、shadow speak（シャドーイング）の素材として完璧です。話者の語調やポーズがリアルで、日常的な会話に近いため、初心者でも取り組みやすいです。特に「waste time」「spin your wheels」などのイディオムが使われていますが、文脈から意味が推測できるので、学習効率が上がります。IELTS スピーキング対策にも最適で、論理的な話し方や説明の仕方を学べます。

自然な表現を分解してみよう

「Most people who try to learn machine learning quit.」：主語が長くても、「who」以下を軽く読むことで流れがスムーズ。日常会話でも長い主語を使う場合、このように読み方を調整すると自然です。
「You learn way more when you're doing hands-on coding.」：「way more」は「はるかに多く」という意味で、口語でよく使われます。「doing hands-on coding」は「実践的なコーディングをする」という表現で、簡潔かつ明確です。
「Don't spend months trying to learn all of it right here.」：「Don't spend ~」の構文はアドバイスをするときに使いやすく、「right here」で「ここで」というニュアンスを加えて、親しみやすさが増します。

効果的な練習ルーチン：英語シャドーイングのステップ

まず、動画を10秒ずつ再生し、すぐにその後に続けて話します（shadow speech）。声の高さや速さ、ポーズを真似ることが重要です。10秒が慣れたら20秒、30秒と延ばしていきましょう。録音機能を使って自分の声を録り、元の動画と比較して、どこが違うかを確認します。「Python fundamentals」「data structures」などの専門用語も、正しい発音を覚えることで自信がつきます。毎日15分、1週間続けると、リズムと流暢さが大幅に向上します。shadowspeakのコツは「繰り返し」です！

この動画の文法

話し手がよく使っている文型を、動画の実際の表現とともに紹介します。

文型	動画での表現
関係詞節 who / which + 節 — 人や物について情報を加える	people who try · one, which is · subject, which is
現在完了形 have/has + 過去分詞 — 過去の出来事が今も関係している	I've seen · have unsupervised · I've partnered
受動態 be + 過去分詞 — 誰がするかより、何が起きるかに焦点を当てる	is supervised · is trusted · be cut

シャドーイングとは？英語上達に効果的な理由

シャドーイング（Shadowing）は、もともとプロの通訳者養成プログラムで開発された言語学習法で、多言語習得者として知られるDr. Alexander Arguelles によって広く普及されました。方法はシンプルですが非常に効果的：ネイティブスピーカーの英語を聞きながら、1〜2秒の遅延で声に出してすぐに繰り返す——まるで「影（shadow）」のように話者を追いかけます。文法ドリルや受動的なリスニングと異なり、シャドーイングは脳と口の筋肉が同時にリアルタイムで英語を処理・再現することを強制します。研究により、発音精度、抑揚、リズム、連音、リスニング力、そして会話の流暢さが大幅に向上することが確認されています。IELTSスピーキング対策や自然な英語コミュニケーションを目指す方に特におすすめです。

シャドーイングのやり方: ステップ別の完全ガイドを読む →

シャドーイング練習: How to learn Machine Learning like a GENIUS and not waste time - 動画で英語スピーキングを学ぶ

Claude Code: The Advanced Guide (99% of Devs Skip These Features)

Learn C++ With Me #19 - Sets

How To Pass Coding Interviews Like the Top 1%

The Art of Learning Like a Genius (even if you're not one)

How I Would Start Cybersecurity in 2026 (If Starting From Zero)

Is ML Engineering Dead? (AI vs ML Engineer)

How I’d Learn Coding in 2026 (Without Wasting Years)

Memory is crazy expensive so I made my own…

8 steps to get your sh** together

It’s Actually Pretty Easy to Get Ahead of 99% of People with AI

This Turns Your Expertise Into a YEAR of Binge-able YouTube Videos

11. Introduction to Machine Learning

If you’re an aspiring product designer, watch this (roadmap included)

How To Learn Any New Skill So Fast It’s Unfair

I cloned 3 apps and now make $35K/month

この動画がスピーキング練習に最適な理由

自然な表現を分解してみよう

効果的な練習ルーチン：英語シャドーイングのステップ

この動画の文法

シャドーイングとは？英語上達に効果的な理由