Machine Learning Zoomcamp — Machine Learning Engineering from Model to Production

Topics covered

• •Understand what ML is: features, the target variable, and extracting patterns from data.
• •Distinguish the ML approach from rule-based systems and know when each is appropriate.
• •Learn the CRISP-DM methodology and the model selection process (train/val/test).

Topics covered

• •Go through the full cycle of a regression ML project: data → EDA → validation → model → tuning.
• •Understand linear regression in scalar and vector form and train it via the normal equation.
• •Learn RMSE, feature engineering, one-hot encoding of categories, and regularization.

Topics covered

• •Build a binary classification project: predicting customer churn for a telecom company.
• •Learn feature importance analysis: churn rate, risk ratio, mutual information, correlation.
• •Understand logistic regression and the sigmoid function; train a model in Scikit-learn.

Topics covered

• •Understand why accuracy is misleading with imbalanced classes.
• •Learn the confusion matrix, precision, recall, and choosing a classification threshold.
• •Understand the ROC curve and the AUC metric, and be able to build and read them.

Topics covered

• •Save and load trained models (pickle) and turn a notebook into scripts.
• •Wrap a model in a web service (Flask; in the current workshop — FastAPI + uv).
• •Isolate dependencies (Pipenv/uv) and package the service into a Docker image.

Topics covered

• •Build a credit scoring project: predicting borrower default.
• •Understand how a decision tree is trained and why, without constraints, it overfits.
• •Learn random forests and gradient boosting (XGBoost) and their key hyperparameters.

Topics covered

• •Build a clothing image classifier with Keras/TensorFlow (and optionally PyTorch).
• •Understand how convolutional networks are structured: convolutional and fully connected layers.
• •Learn transfer learning based on pretrained models (Xception/ImageNet).

Topics covered

• •Deploy a deep learning model the serverless way: AWS Lambda + Docker + API Gateway.
• •Understand lightweight inference formats (TensorFlow Lite, ONNX) and why they're needed.
• •Deploy models to Kubernetes: TF Serving, gRPC, deployments and services, EKS.