Learning about evaluation metrics in #MLZoomcamp, and this was an eye-opener: an 80% accuracy score can be junk.

If your dataset is 72% "no-churn," a model that just always says "no-churn" gets 72% accuracy. Your "fancy" model might just be slightly better than a dumb baseline! 🫠

#DataTalksClub

What's a single, powerful metric for model comparison? AUC (Area Under the Curve).

It measures a model's ability to separate positive and negative classes, independent of any specific threshold.

🎯 1.0 = Perfect classifier 🤷 0.5 = Random guessing

#MLZoomcamp #DataTalksClub #LearningInPublic

How do you know how stable your model is? K-Fold Cross-Validation.

1. Split data into K folds.
2. Train on K-1, test on 1.
3.Repeat K times.

Get the mean & standard deviation of the scores.

A low std dev means your model's performance is consistent! 😌 #MLZoomcamp #DataTalksClub #LearningInPublic

How does a model predict customer churn? 🤔 With Logistic Regression!

A linear model scores a customer, then the sigmoid function squashes that score into a 0-1 probability. High probability = high churn risk! 🎯

Learn more from Alexey Grigorev!
#MLZoomcamp #DataTalksClub #MachineLearning

This week's #MLZoomcamp project is so practical! We're addressing customer churn by building a model to predict who is likely to leave a telecom service and offer them discounts. It's a classic binary classification problem with a clear business outcome. 🎯 #DataTalksClub #BusinessAnalytics

Data distribution matters! My car price data had a long tail, which is bad news for ML models. A logarithmic transformation was the key to normalizing it. Never underestimate the power of solid data prep! 📊 #MLZoomcamp #DataTalksClub #DataScience

This week on #MLZoomcamp a car price prediction model! 🚗Built a linear regression model 📈 from scratch to predict vehicle MSRP. From data cleaning and EDA to regularization, it was an incredible learning journey. #DataTalksClub #LearningInPublic #Regression #MachineLearning

ML projects need a plan! Learning the CRISP-DM lifecycle:
📈 Business Goal -> 📊 Data -> 🧹 Prep -> 🤖 Model -> ✅ Evaluate -> 🚀 Deploy.

It's a cycle, not a straight line! Loving this structured approach.
#MachineLearning #Process #MLZoomcamp #DataTalksClub

Machine Learning Zoomcamp journey! 🤖 The core concept? Let the machine find the patterns in the data. We provide the examples, it learns the rules. Mind-blowing stuff! 🤯 #MachineLearning #MLZoomcamp #DataTalksClub

Finished Module 3 of the #LLMZoomcamp 🚀

We covered the entire offline evaluation pipeline: from generating a ground truth dataset to assessing performance with metrics like MRR, Hit Rate, and Cosine Similarity, plus the fascinating LLM-as-a-Judge approach.

#LLMZoomcamp #RAG #LLM #Evaluation