- What is it?
- Who is this course for?
- What is the format?
- What makes this course unique?
- Why constrain to open source tools?
- What are my qualifications?
- Why is this free?
- What are the prerequisites?
https://t.co/xmMm9XGK9j
ЁЯФе Putting ML in Production! We're going to publicly develop @madewithml's first ML service. Here is the broad curriculum:
- ЁЯУж Product
- ЁЯФв Data
- ЁЯдЦ Modeling
- ЁЯУЭ Scripting
- ЁЯЫа API
- ЁЯЪА Production
More details (lessons, task, etc.) here: https://t.co/xmMm9XGK9j
Thread ЁЯСЗ
- What is it?
- Who is this course for?
- What is the format?
- What makes this course unique?
- Why constrain to open source tools?
- What are my qualifications?
- Why is this free?
- What are the prerequisites?
https://t.co/xmMm9XGK9j
Putting ML in Production: a guide and code-driven case study on MLOps. We will be developing and deploying Made With ML's first ML service, from Product тЖТ ML тЖТ Production, with open source tools.
- ML developers looking to become end-to-end ML developers.
- Software engineers looking to learn how to responsibly deploy and monitor ML systems.
- Product managers who want to have a comprehensive understanding of the different stages of ML dev.
- Intuition: high level overview of the concepts.
- Code: simple code examples to illustrate the concept.
- Application: applying the concept to our specific task.
- Extensions: brief look at other tools and techniques that will be useful.
1. Hands-on
2. Intuition-first
3. Software engineering
4. Focused yet holistic
5. Open source tools
If you search production ML or MLOps online, you'll find great blog posts and tweets. But in order to really understand these concepts, you need to implement them.
We will never jump straight to code. In every lesson, we will develop intuition for the concepts and think about it from a product perspective.
This course isn't just about ML. In fact, it's mostly about clean software engineering! We'll cover important concepts like versioning, testing, logging, etc. that really makes this a production-grade product.
For every concept, we'll not only cover what's most important for our specific task (this is the case study aspect) but we'll also cover related methods (this is the guide aspect) which may prove to be useful in other situations.
We will be using only open source tools for this project, with the exception of @googlecloud for storage and compute (free credit will be plenty).
1. We can focus on the fundamentals, everyone can participate (single player mode as my friend @eugeneyan coined) and you will have much better understanding when you do finally use a paid tool at work (if you want to).
1. I've deployed large scale ML systems at @Apple as well as smaller systems with constraints at startups and want to share the common principles I've learned along the way.
LinkedIn: https://t.co/xWmPKz53vw
Personal website: https://t.co/NpLSczadPn
1. Personal reason: Every day, people explore the amazing work on @madewithml to learn from and contribute themselves. To stay consistent with this free spirit, I'm releasing this free course to pass on the lessons I've learned from my mentors and experiences.
- You should have some familiarity with Python and basic ML algorithms. While we will be experimenting with deep learning (w.r.t compute/performance tradeoffs), you can easily apply the lessons to any class of ML models.
https://t.co/V35zXocadQ
https://t.co/cmTTeALWz1
More from Data science
Wellll... A few weeks back I started working on a tutorial for our lab's Code Club on how to make shitty graphs. It was too dispiriting and I balked. A twitter workshop with figures and code:
Here's the code to generate the data frame. You can get the "raw" data from https://t.co/jcTE5t0uBT
Obligatory stacked bar chart that hides any sense of variation in the data
Obligatory stacked bar chart that shows all the things and yet shows absolutely nothing at the same time
STACKED Donut plot. Who doesn't want a donut? Who wouldn't want a stack of them!?! This took forever to render and looked worse than it should because coord_polar doesn't do scales="free_x".
When are you doing pie charts?
— #BlackLivesMatter (@surt_lab) October 13, 2020
Here's the code to generate the data frame. You can get the "raw" data from https://t.co/jcTE5t0uBT
Obligatory stacked bar chart that hides any sense of variation in the data
Obligatory stacked bar chart that shows all the things and yet shows absolutely nothing at the same time
STACKED Donut plot. Who doesn't want a donut? Who wouldn't want a stack of them!?! This took forever to render and looked worse than it should because coord_polar doesn't do scales="free_x".
To my JVM friends looking to explore Machine Learning techniques - you donтАЩt necessarily have to learn Python to do that. There are libraries you can use from the comfort of your JVM environment. ЁЯз╡ЁЯСЗ
https://t.co/EwwOzgfDca : Deep Learning framework in Java that supports the whole cycle: from data loading and preprocessing to building and tuning a variety deep learning networks.
https://t.co/J4qMzPAZ6u Framework for defining machine learning models, including feature generation and transformations, as directed acyclic graphs (DAGs).
https://t.co/9IgKkSxPCq a machine learning library in Java that provides multi-class classification, regression, clustering, anomaly detection and multi-label classification.
https://t.co/EAqn2YngIE : TensorFlow Java API (experimental)
https://t.co/EwwOzgfDca : Deep Learning framework in Java that supports the whole cycle: from data loading and preprocessing to building and tuning a variety deep learning networks.
https://t.co/J4qMzPAZ6u Framework for defining machine learning models, including feature generation and transformations, as directed acyclic graphs (DAGs).
https://t.co/9IgKkSxPCq a machine learning library in Java that provides multi-class classification, regression, clustering, anomaly detection and multi-label classification.
https://t.co/EAqn2YngIE : TensorFlow Java API (experimental)
1/ A тИЮ-wide NN of *any architecture* is a Gaussian process (GP) at init. The NN in fact evolves linearly in function space under SGD, so is a GP at *any time* during training. https://t.co/v1b6kndqCk With Tensor Programs, we can calculate this time-evolving GP w/o trainin any NN
2/ In this gif, narrow relu networks have high probability of initializing near the 0 function (because of relu) and getting stuck. This causes the function distribution to become multi-modal over time. However, for wide relu networks this is not an issue.
3/ This time-evolving GP depends on two kernels: the kernel describing the GP at init, and the kernel describing the linear evolution of this GP. The former is the NNGP kernel, and the latter is the Neural Tangent Kernel (NTK).
4/ Once we have these two kernels, we can derive the GP mean and covariance at any time t via straightforward linear algebra.
5/ So it remains to calculate the NNGP kernel and NT kernel for any given architecture. The first is described in https://t.co/cFWfNC5ALC and in this thread
2/ In this gif, narrow relu networks have high probability of initializing near the 0 function (because of relu) and getting stuck. This causes the function distribution to become multi-modal over time. However, for wide relu networks this is not an issue.
3/ This time-evolving GP depends on two kernels: the kernel describing the GP at init, and the kernel describing the linear evolution of this GP. The former is the NNGP kernel, and the latter is the Neural Tangent Kernel (NTK).
4/ Once we have these two kernels, we can derive the GP mean and covariance at any time t via straightforward linear algebra.
5/ So it remains to calculate the NNGP kernel and NT kernel for any given architecture. The first is described in https://t.co/cFWfNC5ALC and in this thread
