🎓 Finished Machine Learning Zoomcamp by DataTalks.Club
4 months of ML engineering from scratch → production
• Built models from ground up (regression, trees, neural nets)
• Deployed w/ Docker, Kubernetes, AWS Lambda
• PyTorch, TensorFlow, FastAPI
Shoutout to the DTC
#MLZoomcamp #MachineLearning
🎉 Capstone 2 complete #mlzoomcamp 2025
Human Activity Recognition using smartphone sensors:
📱 95.5% accuracy with Logistic Regression
🚶 6 activities: Walking, Sitting, Standing, Laying...
🐳 Docker + Kubernetes deployed
Key insight: Simple models beat Neural Networks
smplu.link/MWEEh
Reviewed an ML Zoomcamp capstone that had:
✓ SMOTE for 2.21% fraud imbalance
✓ Multi-stage Docker build
✓ Live Render deployment
✓ Custom web UI with risk gauge
16/16. This is how you stand out.
#mlzoomcamp #machinelearning
Peer reviewing ML projects hits different when someone builds a full interactive web UI on top of their fraud detection API 🔥
Not required. Just... done right.
That's the gap between "completed the assignment" and "portfolio-ready."
#mlzoomcamp #datascience
Just reviewed a 16/16 perfect score Capstone for ML Zoomcamp 2025 🏆
E-commerce fraud detection with:
- 5 models compared (XGBoost + SMOTE won)
- 17 → 56 engineered features
- Interactive web UI
- Live cloud deployment
This is what "above and beyond" looks like.
#mlzoomcamp
TIL reviewing someone else's ML project teaches you more than you'd expect 🧠
Just wrapped up a Capstone 1 peer review for ML Zoomcamp — solid depression prediction project with Flask API + Docker deployment.
Peer review > passive learning
#mlzoomcamp
Key insight: A good README and working Dockerfile matter as much as model accuracy. Reproducibility is a skill.
Take peer reviews seriously — you learn just as much evaluating as building.
#mlzoomcamp #datascience
Just finished my first peer review for ML Zoomcamp 2025 Capstone 1 🔍
Reviewing others' projects = underrated learning hack. You see different approaches, pick up documentation tricks, and really appreciate what makes code reproducible.
#mlzoomcamp #machinelearning
🎉 Capstone Project DONE for #mlzoomcamp 2025!
Built an end-to-end Flower Classification system using Transfer Learning 🌸
Here's what I learned 👇
Another module in the books! This time experimenting with serverless deployment using AWS Lambda. Scikit-learn, Keras and Pytorch models, all implemented with Lambda functions, pushed as container images to AWS ECR, and created within AWS Lambda #MLzoomcamp
Module 10 #mlzoomcamp
Learned
🔹 Deploying ML models to Kubernetes with kind
🔹 Pods, Deployments & Services
🔹 Writing YAML configs
🔹 Horizontal Pod Autoscaler (HPA)
✨ Watched ML service auto-scale from 1→3 replicas under load
github.com/HighviewOne/machine-learning-zoomcamp-homework/tree/main/HW10
Wrap up Module 9 #mlzoomcamp 🚀
Learned
🔹 Serverless deployment w AWS Lambda
🔹 ONNX model conversion & inference
🔹 Docker containers for Lambda
🔹 Building & testing locally before deploying
✨ Takeaway: Preprocessing must match training — wrong normalization = wrong predictions
➡️ Module 10 Kubernetes
Wrapped Module 8 #mlzoomcamp 🚀
🔹 Building CNNs from scratch w/ PyTorch
🔹 BCEWithLogitsLoss for binary classification
🔹 Data augmentation (rotation, crop, flip)
🔹 Model architecture: 20M+ parameters!
✨ Takeaway: Data augmentation boosted validation accuracy from 71% → 75%. Small changes, big impact!
Been quietly working on this, limited time with my PC out of commission for a week, but we got there.
Midterm Project - containerised and deployed!
Predicting Type II Diabetes in Women who with history of Gestational Diabetes.
#mlzoomcamp @DataTalksClub
Just wrapped up Module 6 of #mlzoomcamp
Learned about
🔹 Decision Trees for regression
🔹 Random Forest
🔹 Feature importance
🔹 XGBoost parameter tuning
✨More trees isn't always better. There's a sweet spot for model complexity vs performance
➡️ Next up: Midterm project
@Al_Grigor #DataTalkClub
Another module in the books in #mlzoomcamp. This module centered on decision trees and ensemble methods such as Random Forests and XGBoost! It is so easy to over fit a decision tree!
Week5 of revisiting ML concepts: FastAPI > Flask — mainly thanks to Pydantic for clean validation and schema handling.
If you’re just learning this: it will get easier. Practice turns confusion into confidence. 💪 github.com/DataTalksClub
#DataScience #ML #MLOps #MLzoomcamp #LearningInPublic
Docker was key in this weeks deployment week of #mlzoomcamp Made all the more interesting my file structure choices. I faced a few challenges in this one...
Just wrapped up Module 5 of #mlzoomcamp
Learned about:
🔹 Deploying ML models as web services
🔹 Serving predictions with flask
🔹 Dependency mgt with Pipenv
🔹 Dockerizing Python apps
✨ Takeaway: ML models are only useful when they’re live
➡️ Next: Deploy in AWS ebs
@Al_Grigor @DataTalksClub
The Confusion Matrix is your friend. It tabulates the True Positives, True Negatives, False Positives, and False Negatives, giving you the raw numbers behind your model's performance. Start here to truly understand your classifier's errors
#mlzoomcamp
Don't trust accuracy alone! For imbalanced datasets, a model predicting the majority class will have high accuracy but low utility. Always use other metrics like precision, recall, and F1-score for a full picture
#mlzoomcamp
For multi-class problems, you can generate a classification report that provides precision, recall, and F1-scores for each individual class, or averaged metrics like macro and weighted averages
#mlzoomcamp
The ROC curve plots the True Positive Rate vs. the False Positive Rate at different classification thresholds. It visualizes the trade-off between sensitivity and specificity
#mlzoomcamp
F1-score is the harmonic mean of precision and recall. It gives you a single metric that balances both, making it especially useful for evaluating models on imbalanced data
#mlzoomcamp
Precision vs. Recall: The eternal balancing act. Prioritize precision to minimize false positives (e.g., spam detection), or recall to minimize false negatives (e.g., disease diagnosis)
#mlzoomcamp
"Accuracy isn't the whole story! For imbalanced datasets, a high accuracy might be misleading. Always check your confusion matrix to see what’s really going on under the hood
#mlzoomcamp
ML is immense field and intimidating at times. I am trying to make sense of it step by step.
#mlzoomcamp
I think I understand why is MLOps important? Without it, ML models face chaos in prod: degrading performance, manual deployment issues, and scaling challenges. MLOps creates a smooth, automated pipeline for continuous monitoring, retraining
Hope Alexey keeps the #MLOps course alive
#mlzoomcamp
ASGI Bandwagon - An ASGI (Asynchronous Server Gateway Interface) server is a web server that runs asynchronous Python web apps by implementing the ASGI spec. It is a modern, high-perf, asynchronous web frameworks, allowing handling of multiple requests concurrently without blocking.
#mlzoomcamp
