A practical, self-paced course covering machine learning from scikit-learn fundamentals to gradient boosting and deep learning with PyTorch.
Three Jupyter notebooks combine explanations, runnable examples, and hands-on exercises. The focus is on understanding models, evaluating them correctly, and building practical skills.
The course notebooks are written in Ukrainian.
This course is for learners who know basic Python and want a structured introduction to machine learning.
Recommended background:
- Python functions, loops, and basic classes;
- basic NumPy and pandas operations;
- introductory statistics: means, variance, and probability.
No previous experience with neural networks is required.
| Part | Topics | Notebook |
|---|---|---|
| 1. ML Fundamentals | Feature types, train/test splits, regression, classification, metrics, cross-validation, pipelines, regularization, trees, clustering, and an introduction to deep learning | Open notebook |
| 2. Tabular ML and Boosting | Feature engineering, XGBoost, LightGBM, CatBoost, early stopping, interpretation, calibration, ensembles, and time-based validation | Open notebook |
| 3. PyTorch and Deep Learning | Tensors, autograd, training loops, MLPs, regularization, optimizers, CNNs, saving models, and GPU usage | Open notebook |
Each part includes practical exercises. Empty code cells following the exercises are provided for your solutions.
- Work through the notebooks in order: 1 → 2 → 3.
- Run the cells from top to bottom and inspect the results.
- Change parameters and observe how model behavior changes.
- Complete the exercises independently and record your conclusions.
- Apply what you learn in a separate project using an open dataset.
Pay particular attention to validation, metric selection, and preventing data leakage. These determine whether your model evaluation is reliable.
You will need a Python 3 version supported by the libraries below.
Clone the repository:
git clone https://github.com/EternalSecond/ML-Learn.git
cd ML-LearnCreate a virtual environment:
python -m venv .venvActivate it.
Windows PowerShell:
.\.venv\Scripts\Activate.ps1Linux / macOS:
source .venv/bin/activateInstall the dependencies:
python -m pip install --upgrade pip
python -m pip install jupyterlab numpy pandas matplotlib scipy scikit-learn
python -m pip install xgboost lightgbm catboost
python -m pip install torch torchvisionFor the optional SHAP examples in Part 2:
python -m pip install shapStart JupyterLab:
jupyter labOpen a notebook and select the Python kernel with the installed dependencies.
A CPU is sufficient for the course examples. For NVIDIA GPU support, use the installation command recommended for your system on the PyTorch website.
You can view the notebooks directly on GitHub without installing Python.
To run them in Google Colab, open the GitHub tab in Colab or upload a notebook manually. Install any additional dependencies in the Colab environment as needed.
Suggested project directions include:
- classification and regression on open tabular datasets;
- customer churn prediction;
- demand forecasting with time-based validation;
- integrating an ML model into a web application;
- image classification and further exploration of transfer learning.
For each completed project, document the problem, data, validation strategy, metrics, results, and model limitations.
This course was created by EternalSecond, with assistance from Claude (Anthropic) in preparing some of the materials.
This is an independent educational project. It is not an official course from Anthropic or the developers of the libraries used.
Found an error, an unclear explanation, or a technical issue? Open an Issue and include the notebook name and relevant section.
Corrections and improvements to explanations, exercises, and code examples are welcome through Pull Requests.
Textual explanations and educational content are licensed under Creative Commons Attribution 4.0 International (CC BY 4.0).
When sharing the original materials or an adaptation, you must credit EternalSecond, link to the source and license, and indicate any changes. You may not present the original course materials as your own work.
Code in the notebooks is licensed under Apache License 2.0. See LICENSE for the full terms.
Third-party libraries, datasets, and external materials remain subject to their own licenses and terms.