Skip to content

Latest commit

 

History

History
336 lines (225 loc) · 30.4 KB

README.md

File metadata and controls

336 lines (225 loc) · 30.4 KB
Raw

Binder

Introduction to Machine Learning

While most AI research focuses on applying deep learning to unstructured data such as text and images, many real-world AI applications involve applying machine learning to structured, tabular data. This course's objective is to provide students with practical, hands-on experience with state-of-the-art machine learning tools widely used in industry to solve science and engineering problems based on structured, tabular data.

On completion of this course students will:

  • Implement complete machine learning pipelines using Scikit-Learn to solve supervised and unsupervised learning problems with a variety of techniques;
  • Understand various failure modes that can arise when training machine learning pipelines and be able to recognize and troubleshoot these failure modes in their own work;
  • Develop, train, and deploy complete machine learning applications.This course covers the basic theory behind ML algorithms but the majority of the focus is on hands-on examples using Scikit Learn.

Learning Objectives

Capabilities (knowledge & understanding):

  • Understand the distinct types of machine learning problems that exist in practice; understand the basic parts of typical machine learning solutions.
  • Understand the limitations of linear models; understand the trade-offs of various non-linear models.
  • Understand failure modes that are encountered when building machine learning pipelines.

Competencies (skills):

  • Be able to implement a complete machine learning workflow using Scikit-Learn to solve supervised and unsupervised learning problems using linear models.
  • Train, validate, and test machine learning pipelines using non-linear models with Scikit Learn.
  • Analyze and ‘troubleshoot’ failure modes in their own work.

Lessons

Module 1: Introduction to Machine Learning, Part I

Lecture 1: The Machine Learning Landscape

The initial lecture provides a high-level overview of the machine learning (ML) landscape and addresses the following questions.

  • What is ML?
  • Why use ML?
  • What are the several types of ML systems?
  • What are some of the main challenges of applying ML in practice?

The material covered in this lecture is drawn from the following chapters of the reference texts.

Lecture 2: End-to-End Machine Learning Project

In this lecture, we will define and discuss the steps for any applied ML project by working through an end-to-end ML project using a real-world dataset as a group.

  1. Framing the problem
  2. Getting the data
  3. Exploratory data analysis
  4. Preparing the data for ML algorithms
  5. Selecting promising ML algorithms and training
  6. Tuning ML algorithms for performance
  7. Deploying tuned ML algorithms

The goal is to provide students with a roadmap or checklist that they can apply to most any ML project. Subsequent lectures will focus on various steps defined above.

The material covered in this lecture is drawn from the following chapters of the reference texts.

Tutorial Open in Google Colab Open in Kaggle
Get the Data Google Colab Kaggle
Exploratory Analysis Google Colab Kaggle
Preparing Data for ML Google Colab Kaggle
Training Models Google Colab Kaggle
Fine-tuning Models Google Colab Kaggle

Lecture 3: Classification

In this lecture we will cover the key ideas of one of the major learning tasks performed by supervised ML systems: classification.

  • Differentiate between several types of classification: binary and multi-class classification; multi-output and multi-label classification; soft vs hard classification.
  • Selecting an appropriate performance metric is a critical early step in developing a classification system.
  • Discuss the importance of calibrating predicted class probabilities.

The material covered in this lecture is drawn from the following chapters of the reference texts.

  • Chapter 3 of Hands-on ML with Sklearn, Keras, and TF
  • Chapter 2 of ML with PyTorch and Sklearn
Tutorial Open in Google Colab Open in Kaggle
Binary Classification Google Colab Kaggle
Performance Metrics Google Colab Kaggle
Multiclass Classification Google Colab Kaggle
Multilabel Classification Google Colab Kaggle
Multioutput Classification Google Colab Kaggle
Calibration Google Colab Kaggle
End-to-end Classification Google Colab Kaggle

Lecture 4: Training Models, part I

This lecture provides an in-depth discussion of the process of training ML models using stochastic gradient descent. The process will be illustrated using various linear regression models.

  • Linear regression
  • Polynomial regression
  • Regularized linear regression (Ridge, LASSO, and ElasticNet).

The material covered in this lecture is drawn from the following chapters of the reference texts.

Tutorial Open in Google Colab Open in Kaggle
Linear Regression with NumPy Google Colab Kaggle
Linear Regression with Scikit-Learn Google Colab Kaggle
Data Preprocessing Google Colab Kaggle
Polynomial Regression Google Colab Kaggle
Regularized Linear Regression Google Colab Kaggle

Lecture 5: Training Models, part II

Review the key ideas presented in the previous lecture but this time using logistic and softmax regression (with and without regularization) to illustrate the training process.

The material covered in this lecture is drawn from the following chapters of the reference texts.

Tutorial Open in Google Colab Open in Kaggle
Logistic Regression with Numpy Google Colab Kaggle
Softmax Regression with Numpy Google Colab Kaggle
Logistic Regression with Scikit-Learn Google Colab Kaggle
Softmax Regression with Scikit-Learn Google Colab Kaggle

Module 2: Introduction to Machine Learning, Part II

Lecture 1: Support Vector Machines (SVMs)

In this lecture we will cover both linear and non-linear support vector machines (SVMs) and see how to apply these algorithms to solve both classification and regression tasks.

The material covered in this lecture is drawn from the following chapters of the reference texts.

Tutorial Open in Google Colab Open in Kaggle
SVMs (Classification) Google Colab Kaggle

Lecture 2: Decision Trees

Tree-based models such as random forests and gradient-boosted trees are state-of-the-art ML methods for tabular data. But before we can cover these state-of-the-art methods, we need to discuss decision trees in detail. This lecture will focus on the key ideas behind decision tree implementation and how to tune decision trees to avoid overfitting.

The material covered in this lecture is drawn from the following chapters of the reference texts.

Tutorial Open in Google Colab Open in Kaggle
Decision Trees (Classification) Google Colab Kaggle
Decision Trees (Regression) Google Colab Kaggle

Lecture 3: Ensemble Methods, part I

Ensemble, tree-based models such as random forests and gradient-boosted trees are state-of-the-art ML methods for tabular data. In this lecture we will cover these approaches in detail. We will discuss the key ideas behind ensemble methods such as voting, bagging, and pasting and how to implement these approaches in practice to solve classification and regression tasks via random forests.

The material covered in this lecture is drawn from the following chapters of the reference texts.

Tutorial Open in Google Colab Open in Kaggle
Ensemble Methods, part I Google Colab Kaggle

Lecture 4: Ensemble Methods, part II

This lecture continues covering tree-based ensemble techniques with an in-depth discussion of gradient boosted trees. Gradient boosted trees are the most widely used ML algorithms in industry and continue to deliver state-of-the-art performance on tabular data problems and time-series forecasting problems. Two popular implementations of gradient boosted trees will be discussed: XGBoost and CatBoost.

The material covered in this lecture is drawn from the following chapters of the reference texts.

Tutorial Open in Google Colab Open in Kaggle
Boosting Methods (Scikit-Learn) Google Colab Kaggle
Boosting Methods (CatBoost, LightGBM, XGBoost) Google Colab Kaggle

Module 3: Introduction to Machine Learning, Part III

Lecture 1: Dimensionality Reduction

Many ML systems suffer from the curse of dimensionality: the training time of many ML systems increases rapidly as the number of features in the dataset increases. This lecture covers the key ideas of dimensionality reduction and various algorithms for performing dimensionality reduction such as Principal Components Analysis (PCA), Locally Linear Embedding (LLE), et al. Tradeoffs between the various methods will be discussed.

The material covered in this lecture is drawn from the following chapters of the reference texts.

Tutorial Open in Google Colab Open in Kaggle
Dimensionality Reduction Google Colab Kaggle

Lecture 2: Unsupervised Learning Techniques

This lecture will cover the key ideas behind unsupervised learning techniques and will include an in-depth discussion of the K-means algorithm. Other approaches to unsupervised learning will be discussed such as DBSCAN, and Gaussian Mixture Models (GMMs).

The material covered in this lecture is drawn from the following chapters of the reference texts.

Tutorial Open in Google Colab Open in Kaggle
Unsupervised Learning Google Colab Kaggle

Module 4: Introduction to Machine Learning, Part IV

Lecture 1: Introduction to Artificial Neural Networks

While neural networks are not (yet?) state-of-the-art approach for most tabular data problems, an understanding of the key ideas of basic neural networks such as the multi-layer perceptron (MLP) is important. We will learn how to implement linear and logistic regression using MLPs as well as how to apply MLPs to tabular data problems.

The material covered in this lecture is drawn from the following chapters of the reference texts.

Lecture 2: Fundamentals for DNN training and validation

The material covered in this lecture is drawn from the following chapters of the reference texts.

Lecture 3: Optimizers, learning rates and batch sizes

The material covered in this lecture is drawn from the following chapters of the reference texts.

Module 5:

Lecture 1: Deploying machine learning pipelines, part 1

A trained ML system that is never deployed is one of little practical value. Deployment of ML systems presents some unique challenges that are rarely discussed in university courses. We will spend the last lecture of the course learning how to deploy ML algorithms culminating with students learning to embed and deploy their own ML systems as Gradio applications on HuggingFace Spaces.

Assessment

Student performance on the course will be assessed through participation in a Kaggle classroom competition.

Repository Organization

Repository organization is based on ideas from Good Enough Practices for Scientific Computing.

  1. Put each project in its own directory, which is named after the project.
  2. Put external scripts or compiled programs in the bin directory.
  3. Put raw data and metadata in a data directory.
  4. Put text documents associated with the project in the doc directory.
  5. Put all Docker related files in the docker directory.
  6. Install the Conda environment into an env directory.
  7. Put all notebooks in the notebooks directory.
  8. Put files generated during cleanup and analysis in a results directory.
  9. Put project source code in the src directory.
  10. Name all files to reflect their content or function.

Building the Conda environment

After adding any necessary dependencies that should be downloaded via conda to the environment.yml file and any dependencies that should be downloaded via pip to the requirements.txt file you create the Conda environment in a sub-directory ./envof your project directory by running the following commands.

export ENV_PREFIX=$PWD/env
mamba env create --prefix $ENV_PREFIX --file environment.yml --force

Once the new environment has been created you can activate the environment with the following command.

conda activate $ENV_PREFIX

Note that the ENV_PREFIX directory is not under version control as it can always be re-created as necessary.

For your convenience these commands have been combined in a shell script ./bin/create-conda-env.sh. Running the shell script will create the Conda environment, activate the Conda environment, and build JupyterLab with any additional extensions. The script should be run from the project root directory as follows.

./bin/create-conda-env.sh

Ibex

The most efficient way to build Conda environments on Ibex is to launch the environment creation script as a job on the debug partition via Slurm. For your convenience a Slurm job script ./bin/create-conda-env.sbatch is included. The script should be run from the project root directory as follows.

sbatch ./bin/create-conda-env.sbatch

Listing the full contents of the Conda environment

The list of explicit dependencies for the project are listed in the environment.yml file. To see the full lost of packages installed into the environment run the following command.

conda list --prefix $ENV_PREFIX

Updating the Conda environment

If you add (remove) dependencies to (from) the environment.yml file or the requirements.txt file after the environment has already been created, then you can re-create the environment with the following command.

$ mamba env create --prefix $ENV_PREFIX --file environment.yml --force

Using Docker

In order to build Docker images for your project and run containers with GPU acceleration you will need to install Docker, Docker Compose and the NVIDIA Docker runtime.

Detailed instructions for using Docker to build and image and launch containers can be found in the docker/README.md.