Santiago Profile picture
16 Feb, 7 tweets, 2 min read
Imagine that you ask a yes/no question to 1,000 people, and each person answers correctly 51% of the time.

You count the different answers and pick the most common one.

How likely are you to end up with the correct answer?

🧵👇
Don't feel bad if you think it's 51% — We all did!

If every person answers independently from the rest, you'll end up with the correct answer ~75% of the time.

And if you ask 10,000 people, the chance of getting it right goes up to ~97%!

Mind-blowing, right?

(2 / 6)
If you care about the math behind this, take a look at the attached expression. (But it doesn't matter if you don't.)

This is what's important:

The law of large numbers ensures that we get more correct answers as we ask more people.

(3 / 6)
Let's imagine now that, instead of people, we have many machine learning models.

If we compute our final prediction by combining all the models' answers, we will have a better chance of getting the correct result!

This illustrates the power of an ensemble model.

(4 / 6)
Keep in mind that the numbers here only work if each model's prediction is independent of every other model's prediction.

This is unlikely because models will be looking at the same data.

However, combining multiple models is still much better 😎.

(5 / 6)
If you are curious, we call models that do only slightly better than random guessing "weak learners."

If you combine a sufficient number of weak learners together in an ensemble, you'll get a "strong learner," which will give you much better results.

(6 / 6)
If you enjoy my attempts to make machine learning a little more intuitive, stay tuned and check out @svpino for more of these threads.

• • •

Missing some Tweet in this thread? You can try to force a refresh
 

Keep Current with Santiago

Santiago Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!

PDF

Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

More from @svpino

15 Feb
I have not seen any proof that Twitter "kills your content" if you include links to your tweets.

Here is the result of a very unscientific experiment: comparing my top 10 tweets with and without links.

If you have something concrete, please let me know.
This is anecdotal evidence at best.

It doesn't prove that Twitter doesn't mess with your links, but it does suggest that —if anything is going on— it is much more subtle than what some believe.

I haven't found any documentation either.
This is what I do know:

Breaking the links that you add to your tweets is self-serving: it makes it worse for the people who follow you. They can't just click to get the content.

I can't see how this will make your content better in any way.
Read 4 tweets
14 Feb
A collection of the most interesting threads I've written about machine learning.

🧵 x 🧵
Read 13 tweets
13 Feb
It takes a single picture of an animal for my son to start recognizing it everywhere.

Neural networks aren't as good as we are, but they are good enough to be competitive.

This is a thread about neural networks and bunnies.

🧵👇
A few days ago, I discussed how networks identify patterns and use them to extract meaning from images.

Let's start this thread right from where we ended that conversation.



(2 / 16)
Let's assume we use these four pictures to train a neural network. We tell it that they all contain a bunny 🐇.

Our hope is for the network to learn features that are common to these images.

(3 / 16)
Read 17 tweets
11 Feb
Today let's talk about why we keep "splitting the data" into different sets.

Besides machine learning people being quirky, what else is going on here?

Grab your coffee ☕️, and let's do it!

🧵👇
Imagine you are teaching a class.

Your students are getting ready for the exam, and you give them 100 answered questions, so they prepare.

You now need to design the exam.

What's the best way to evaluate the students?

(2 / 19)
If you evaluate the students on the same questions you gave them to prepare, you'll reward those who just memorized the questions.

That won't give you a good measure of how much they learned.

😑

(3 / 19)
Read 21 tweets
10 Feb
Interviews aren't broken.

A lot of people complain about them, yet few have any experience hiring.

This is a rant full of my own biases and limited perspective —a break from machine learning threads.

🧵👇
Building a team is incredibly hard.

Building a good team is even more challenging.

Building a good, diverse team is a nightmare.

👇
Imagine you are starting a new company and you need a couple of developers.

▫️ Where do you find them?
▫️ How do you know they are any good?
▫️ How much do you pay them?

How do you make somebody come and work for you, a nobody?

👇
Read 25 tweets
9 Feb
Seriously though, how the heck can a computer recognize what's in an image?

Grab a coffee ☕️, and let's talk about one of the core ideas that makes this possible.

(I'll try to stay away from the math, I promise.)

👇
If you are a developer, spend a few minutes trying to think about a way to solve this problem:

→ Given an image, you want to build a function that determines whether it shows a person's face.

2/ Image
It gets overwhelming fast, right?

What are you going to do with all of these pixels?

3/
Read 27 tweets

Did Thread Reader help you today?

Support us! We are indie developers!


This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Too expensive? Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal Become our Patreon

Thank you for your support!

Follow Us on Twitter!