🌲🌲 Applied ML/AI, data science, MLOps | Wife of 1, mom of 2 | Co-Founder and CTO of https://t.co/67wk0TNMlO | Quote: Oliver Wendell Holmes 🕊️
Dec 22, 2022 • 13 tweets • 3 min read
Why do ML projects fail? It’s not usually about technology. It’s about people.
👩🏻👦🏽👨🏼🦲🧑🏾🦳👩🏿🦱👨🏼👱🏾👱🏽♂️👩🏼🦳👨🏿🦳👩🏼🦲🧑🏾🦲🧔🏻♂️
First, let’s talk about successful ML projects. Almost invariably, they bring together a group of people with different but overlapping expertise:
<1/13>
① Domain Experts: Deep knowledge of the domain — e.g., an educators marketers, manufacturers, etc.
② ML Experts: Knows ML development — e.g., a Data Scientist or ML Engineer.
③ Coordination Expert: Knows how to deploy solutions in the domain — e.g., a PM.
<2/13>
Nov 1, 2022 • 4 tweets • 1 min read
📚 ML/Data Science beginners don’t need an encyclopedic cheat sheet of all types of ML models. They don’t need to understand linear algebra, statistics, or calculus.
📄They need to understand the components of a simple, complete ML system.
That means they need to know…
1. What a feature set is, and one example 2. What a model is, and one example 3. What a performance metric is, and one example 4. Example code that combines the three
Then, switch out each of the first 3 one by one to show how those components are modular.
📊Explaining accuracy to a non-technical stakeholder
Too high-level, and they suspect you're hiding something. Too granular, and they'll be lost in the weeds.
My solve? Visualize F1 👇
First, I find F1 navigates simplicity and power. This lets me earn trust by briefly explaining the downside of traditional Accuracy:
"Imagine predicting fraud where only 1% of the transactions were fraudulent. If the model predicted that none were fraudulent, it would be..."
Sep 10, 2022 • 5 tweets • 1 min read
This is why having one makes you irreplaceable! I don’t know of a place to go to find sets of metrics frameworks by function or org type (add a link if you have one!).
But I do have a method for developing them — think in terms of inputs and outputs.
Every organization is in the business of turning inputs into outputs.
Whether in manufacturing where raw materials become products, or in education where applicants become graduates, something goes in one end of an org, gets processed, and comes out changed.
Sep 10, 2022 • 8 tweets • 2 min read
Whether you’re a #DataScientist, #Analyst, or #MachineLearning Engineer, fill your toolbox with one of the most important tools for any job:
🧰 Metrics Frameworks
They’re the shortcut to being effective and irreplaceable.
Example:
Take a org type like an e-commerce business. Most of them have the same functions within them: sales and/or marketing, finance, product, UX (site/app design and maintenance), logistics, customer service, legal, etc.
Seems like a ton of disparate data! Well, turns out…
