Full-Stack Deep Learning with Python

Course details

Full-stack deep learning encompasses the complete lifecycle of building and deploying machine learning systems—from project planning and data preparation to model training, optimization, and deployment. In this course, join data engineer Janani Ravi as she explores each stage of the lifecycle in Python, using MLflow for MLOps and Optuna for hyperparameter tuning. Learn how to manage machine learning artifacts and environments for reproducibility and scalability, and practice deploying models to serve real-world applications. Upon completing this course, you’ll be equipped with the skills you need to automate and optimize machine learning processes and build full-stack deep learning systems from end to end.

Concepts

Introduction

Full-stack landscape and strategy
Full-stack deep learning - MLOps and MLflow
Prerequisites

An Overview of Full-Stack Deep Learning

Components - Planning and data collection
Components - Model training and deployment
Artifacts in full-stack deep learning
Tools - Compute, orchestration, and experiments
Tools - Versioning, labeling, and feature stores
Tools - Deep learning frameworks and debugging
Tools - APIs, UIs, CI CD, and monitoring

MLOps with MLflow

Machine learning operations (MLOps)
Managing the ML lifecycle with MLflow
Setting up the environment on Google Colab
Running MLflow and using ngrok to access the MLflow UI

Model Training and Evaluation Using MLflow

Loading and exploring the EMNIST dataset
Logging metrics parameters and artifacts in MLflow
Set up the dataset and data loader
Configuring the image classification DNN model
Training a model within an MLflow run
Exploring parameters and metrics in MLflow
Making predictions using MLflow artifacts
Preparing data for image classification using CNN
Configuring and training the model using MLflow runs
Visualizing charts metrics and parameters on MLflow

Hyperparameter Tuning with Optuna

Setting up the objective function for hyperparameter tuning
Hyperparameter optimization with Optuna and MLflow
Identifying the best model
Registering a model with the MLflow registry

Model Deployment and Predictions

Setting up MLflow on the local machine
Workaround to get model artifacts on local machine
Deploying and serving the model locally

Conclusion

Summary and next steps