Deep Learning with Python and Pytorch

The course "Deep Learning mit Python und Pytorch" lasts five days. You start with Python and Numpy and build the foundation for machine learning. Then you dive straight into the practical application of deep learning with PyTorch.

At the beginning, you learn Python from scratch: data types, control structures, functions and lambda functions, classes and objects, inheritance, and error handling. After that, Numpy, indexing, and broadcasting are covered. For visual representation, you use Matplotlib. Pandas and Seaborn are briefly introduced in the course.

With this knowledge, you implement a linear regression in Python and Numpy, including gradient descent and loss functions. As an extension, you learn about logistic regression for classification. Then you switch to PyTorch: tensors, their dimensions, data sets, and DataLoaders. Standard datasets like MNIST and FashionMNIST are included.

Finally, you build a Multi Layer Perceptron using nn.Module as the basic building block. You work with loss functions and optimizers and go through a complete training loop. You validate your model with a validation set before making your own predictions.

The course is aimed at developers. Although Python is explained from scratch, the course quickly targets advanced concepts. Therefore, it is not suitable as a general introduction to programming.

You should bring a basic mathematical understanding as well as object-oriented and functional programming skills. Knowledge of a language like JavaScript, Java, C#, or C++ is practical. After participation, you receive a certificate of attendance. Additionally, there is a digital Open Badge that you can share, for example, on LinkedIn.

• Background: historical development of Python and current application areas
• Installation and support in IDEs such as Visual Studio Code and PyCharm
• Jupyter Notebooks as well as cloud offerings such as Google Colab and AWS SageMaker
• Brief tour of the language: data types, control structures, functions, and lambda functions
• Classes and objects, inheritance, error handling, modules, and standard building blocks
• NumPy: application area, data types, basic operations, indexing, and broadcasting
• Visual preparation with Matplotlib, brief introduction to Pandas and Seaborn
• Linear regression: problem statement and implementation with Python and NumPy
• Mathematical background on gradient descent, derivation, and loss functions
• Logistic regression as a variant for classification
• PyTorch: historical development and tensor as basic type with operations
• Detailed treatment of tensor dimensionality
• Linear and logistic regression with PyTorch
• Data set and DataLoader, standard datasets such as MNIST and FashionMNIST
• Multi-layer perceptrons with nn.Module as a basic building block
• Loss functions and optimizers
• Training loop, validation set, and prediction