Tweet

https://twitter.com/1408789941040058369/status/1685921108053172224

This Thread may be Removed Anytime!

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

More from @paulabartabajo_

Pau Labarta Bajo

@paulabartabajo_

Jul 31

How to solve Machine Learning problems in the real world

3 practical tips to make your ML life easier 🧵↓

𝗖𝗼𝗻𝘁𝗲𝘅𝘁

Online courses and Kaggle-style competitions are great resources to learn the fundamentals of ML.

However, the daily job of a machine learning engineer requires an 𝗮𝗱𝗱𝗶𝘁𝗶𝗼𝗻𝗮𝗹 𝗹𝗮𝘆𝗲𝗿 𝗼𝗳 𝘀𝗸𝗶𝗹𝗹𝘀 that you won’t master there.

Here are the 𝘁𝗼𝗽 𝟯 most recurring hidden problems I have faced in my ML life, and my tips for you to deal with them.

Read 18 tweets

Pau Labarta Bajo

@paulabartabajo_

Jul 27

What is 𝗺𝗼𝗱𝗲𝗹 𝗿𝗲-𝘁𝗿𝗮𝗶𝗻𝗶𝗻𝗴 and how to implement it?

Hands-on, in 3 steps🧵↓

𝗧𝗵𝗲 𝗽𝗿𝗼𝗯𝗹𝗲𝗺

No matter how good your predictive Machine Learning model is today, it will eventually expire.

Why?

Because a predictive ML model is essentially a mapping between

→ a set of features (aka inputs)
→ a target (aka output) → what you want to predict

And the thing is, the relationship (aka correlation) between the features and the target can change a lot over time.

This is especially true in problems like recommender systems, or fraud detection.

Read 11 tweets

Pau Labarta Bajo

@paulabartabajo_

Jul 26

XGBoost is one of the most effective algorithms for time-series prediction.

But, you need to prepare your data carefully.

Here is a Python library to help you prepare your data ↓

@joaopcnogueira, one of my students from the Real World ML Tutorial, has built 𝘁𝘀𝟮𝗺𝗹 a Python library that lets you transform

- a time series dataset, into
- a training dataset, with features and targets

Enjoy it

And give it. a star ⭐ on GitHub ↓
github.com/joaopcnogueira…

Wanna build your first real-world ML system?

Join the Real-World ML Tutorial + Community and get LIFETIME ACCESS to

→ 3 hours of video lectures 🎬
→ Full source code 👨‍💻
→ Discord private community 👨‍👩‍👦

Use code "NINJA" at checkout for a 20% discount

realworldmachinelearning.carrd.co

Read 4 tweets

Pau Labarta Bajo

@paulabartabajo_

Jul 26

3 reasons why your XGBoost model does not work

And 3 ways to solve them
↓↓↓

1️⃣ You are overfitting the training data

This is common in highly non-stationary problems, like cryptocurrency price prediction.

Solution. Use cross-validation and hyper-parameter tuning, to adjust the model's bias-variance and get good out-of-sample metrics.

2️⃣ You miss an essential feature in your dataset

You need more/better features to increase the signal-to-noise ratio in your data.

Solution. Pull more raw features and generate better ones through feature engineering.

Read 5 tweets

Pau Labarta Bajo

@paulabartabajo_

Jul 25

Advice for ML beginners💡

GitHub actions are *free* computing that makes your life easier.

Here are 3 use cases for ML projects ↓

➡️ Continuous Integration and Deployment (CI/CD)

Machine Learning is software engineering. As such, it is crucial you automate:

→ code updates (aka integration), and
→ code releases to your production environment (aka deployment)

➡️ Batch feature pipelines

This is a program that runs on a chron-like schedule, that fetches raw data from a data source (e.g. a data warehouse), computes ML features, and saves them to a storage service (e.g. a feature store).

Feature pipelines are present in every ML system.

Read 6 tweets

Pau Labarta Bajo

@paulabartabajo_

Jul 19

Wanna train more ML models for less money? 💸

3 tips to optimize your ML budget 🧠↓

To build a Machine Learning product you need to spend money on 3 types of services:

→ Computing, like CPUs and GPUs so you can train and deploy your models.
→ Orchestration, to kick off the 3 pipelines of your system
→ Storage, to save features, models, and experiment runs

And the thing is, not all these services cost you the same.

→ Orchestration and storage are not expensive 💸
→ Computing, on the other hand, can get very expensive 💸💸💸💸💸

Read 11 tweets

Share this page!

Enter Twitter Thread URL to Unroll

Pau Labarta Bajo

Try unrolling a thread yourself!

More from @paulabartabajo_

Pau Labarta Bajo

Pau Labarta Bajo

Pau Labarta Bajo

Pau Labarta Bajo

Pau Labarta Bajo

Pau Labarta Bajo

Did Thread Reader help you today?

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

Send Email!