Tweet

@thush89

This Thread may be Removed Anytime!

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

More from @svpino

Santiago

@svpino

Aug 10

One of my favorite applications of neural networks: autoencoders.

Here is how they work and how you can use them:

1 of 12

Think of autoencoders as lossy data compression algorithms built using neural networks:

1. A section of the network compresses the input
2. Another section reverses the process

In the middle of these sections, there's a crucial component: a bottleneck.

2 of 12

The bottleneck forces the network to simplify the input data. Anything that's not essential, goes away.

In the end, the bottleneck gives us a compact representation of the dataset.

↑ This is the most important insight here.

3 of 12

Read 14 tweets

Santiago

@svpino

Aug 9

This is a step-by-step guide to building your first deep learning model.

Let's get started:

1/21

Building the model is essential, but it's just the beginning.

I want to give you everything you need to understand what's going on:

• A way to make changes
• A way to experiment
• A way to keep track of everything you did

We are going to use a neat tool for that.

2/21

@Cometml

We are going to use @Cometml to keep track of our experiments.

Create a free account here: comet.com/signup?utm_sou…

You can also connect your GitHub account. The process will take 10 seconds.

3/21

Read 21 tweets

Santiago

@svpino

Aug 5

Here is one of the most spectacular failures in my career.

And it was all my fault:

1 of 12

For the first time, I had the opportunity to lead a decently-sized product.

I was just promoted to Tech Lead.

It was exciting.

2 of 12

Back then, the term "serverless" wasn't a thing.

Google had just come out with App Engine. You could deploy your code, and App Engine will run it and scale it automatically.

I knew Java. App Engine supported Java.

3 of 12

Read 12 tweets

Santiago

@svpino

Aug 2

Overfitting sucks.

But you knew that already.

What you probably don't know is what to do when your model doesn't work on your test set despite doing great on the train set.

I'll show you a clever technique to deal with this:

1 of 13

Here is a common situation:

A team that splits their data based on a specific time frame:

• They grab six months of data
• Train with the first five months
• Test with the final month.

Guess what could happen?

2 of 13

Something may have changed within that timeframe!

If this happened, your test data might differ significantly from your training data.

If train and test don't come from the same distribution, forget about good results.

Fortunately, there's a way to deal with this:

3 of 13

Read 13 tweets

Santiago

@svpino

Jul 29

A lesson some people learn too late:

Building a machine learning system is not like building regular software.

But unless you've done it before, you'd think it's all pretty much the same.

Here is a reason they are different and what you can do about it:

1 of 14

For the most part, regular software looks like this:

• Build once
• Run forever*

I know there's no such thing as *forever*. Things change all the time.

But this rate of change pales in comparison with machine learning systems. Here is an example:

2 of 14

A real, straightforward example:

You build a deep learning model to process pictures that users capture with their phones.

The goal of the model: recognize different shoe brands in the pictures.

Everything works, but something happens:

3 of 14

Read 14 tweets

Santiago

@svpino

Jul 26

Everyone knows they need to replace missing values in their dataset.

Most people, however, miss one critical step.

Here is what you aren't doing and how you can fix it:

1 of 9

I'll start with an example.

A company surveys a bunch of people.

Some people leave one particular question unanswered.

When we collect the data, and before using it to build a model, we must take care of these missing values.

2 of 9

For simplicity's sake, let's assume that replacing missing answers with the mean of all the other answers is a good approach.

We can do that and solve the problem.

Most people stop here. That's their mistake.

3 of 9

Read 12 tweets

Share this page!

Santiago

People who liked this thread also liked...

Try unrolling a thread yourself!

More from @svpino

Santiago

Santiago

Santiago

Santiago

Santiago

Santiago

Did Thread Reader help you today?

Don't want to be a Premium member but still want to support us?