Machine Learning Deployment For Professionals

Explore the production lifecycle of machine learning—from loading data and models to testing and deployment—covering ml ops foundations, TensorFlow Extended, drift monitoring, and catastrophic forgetting mitigations.

Recap the core machine learning pipeline from input data and visualization to training, evaluation, and model saving, and review deployment with TensorFlow serving, TensorFlow Lite, Docker, and TPU hardware.

Explore patterns of machine learning model deployment, from data pipelines and data versioning to model packaging, serving, evaluation, and monitoring, including offline prediction, batch prediction, real time serving, and streaming.

Analyze drift and data shift and apply metrics like L infinity distance and Jensen Shannon divergence to detect changes. Use TensorFlow data validation to monitor schema, skew, and drift.

Identify drift in deployed models and retrain on problematic data points to restore performance. Explore continuous learning and solutions like elastic consolidation with the avalanche library to mitigate catastrophic forgetting.

Identify output data challenges in model serving, including framework and vendor lock-in, training versus serving divergence, and load balancing across cascading models under compute and memory constraints.

Explore building and deploying machine learning projects, from loading datasets and training a simple neural network to saving and serving with TensorFlow, and navigating ML ops and deployment challenges.

Explore the data lifecycle for machine learning, from data modalities and pipelines to storage, cleaning, feature engineering, augmentation, training, testing, and deployment.

Learn the data lifecycle for machine learning deployments, from data sources and exploration to wrangling, train-test splits, versioning, and monitoring, with a focus on modalities and data formats.

Begin manual data collection for production models, detailing timestamps, request id, and metadata, and review computer vision and nlp annotation workflows, exports, and deployment implications.

Explore data preparation with a deployment focus, covering normalization and cleaning techniques across image, text, and tabular data, then dive into anomaly detection with a PyTorch autoencoder and deployment considerations.

explain anomaly detection choices and implement from scratch versus ready-made services, focusing on data preparation, preprocessing, and IBM developer sandbox examples for production-quality deployment.

Explore a Google Cloud production use case detailing data preparation, cohort segmentation, aggregation and ranking, explainability, and deploying anomaly detection models across training and deployment pipelines, with monitoring for auditability.

Explore data wrangling for tabular data using Sage Maker, including feature selection, describe function, correlation plots, confusion matrices, and deployment considerations through visualization and feature stories.

Explore data validation in machine learning deployment, covering train-test splits, cross-validation, anomaly detection, and TensorFlow Data Validation for robust, scalable models.

Explore relational vs NoSQL data stores and the different NoSQL models, including document, key-value, wide-column, and graph stores, with parquet, schema, and desk for machine learning pipelines.

Explore data versioning and data version control as essential tools for tracking data, features, and model iterations with a commit history approach.

Learn feature engineering through data augmentation and test time augmentation for image and text models, with deployment considerations, inference budgets, and model evaluation.

Explore the training pipeline—from model engineering and evaluation to packaging and docker-based serving—highlighting experiment tracking and deployment considerations.

Explore model saving and serving formats such as Onyx, h5, and dot model, and learn to convert between frameworks using Onyx as an intermediate representation, including deploying with SageMaker.

explore monitoring, explainability, and auditability inside a dockerized model workflow using Neptune for visualizations, logging accuracy and loss, and deploying with containerization and model monitoring tools.

Explore the data lifecycle from storage and cleaning to validation and versioning, and apply feature engineering, data augmentation, and test-time augmentation, then cover training, experiment tracking, and deployment.

Explore machine learning pipelines, build automation, and AI model validation, linking quality assurance to software practice, with production case studies, drift monitoring, experiment tracking, data handling, and security considerations.

Gain an overview of machine learning pipelines, from data storage and cleaning to training, monitoring, and deployment with containers, orchestration, and versioning considerations.

Explore quality assurance and validation for machine learning models across the deployment lifecycle, focusing on automation, API model quality, and ongoing evaluation of accuracy, explainability, fairness, and privacy.

Explores how ISO IEC 25010 quality characteristics apply to machine learning models, covering offline and online testing, cross-validation, AB testing, and build automation within CI/CD pipelines.

Explore production ml deployments through case studies of a T-Mobile ar use case and Grubhub pipelines, covering dockerized rest api, semi-supervised learning with Snorkel, and end-to-end build and runtime.

Analyze GrubHub’s production ML pipeline: data extraction and enrichment, weak supervision with Snorkel labeling functions, text classification modeling in TensorFlow, and deployment with saved models and periodic retraining for serving.

Build automation with TensorFlow Extended and Vertex AI pipelines, defining components and triggering re-execution on data or model changes for training, serving, and monitoring.

Map security and privacy to your machine learning pipeline by exploring secure architectures, data lineage, authentication, encryption, and explainability within SageMaker deployments.

Explore the machine learning deployment pipeline, from quality assurance and build automation to secure, privacy-aware workflows on gcp, with case studies in r, docker, snorkel.

Explore how state-of-the-art technologies fit into machine learning deployment lifecycles and pipelines, covering real-time streaming, batch prediction, offline training, shadow environments, edge deployments, and production tools like Flask and Streamlit.

Explore machine learning deployment endpoints, including offline prediction, batch and real-time serving, streaming, cascading models, asynchronous and synchronous pipelines, web browser and embedded machine learning, with regulatory considerations.

Explore Windows and Linux deployment choices for machine learning, comparing development environments, cloud options, and tooling like Docker, WSL, and Azure data science virtual machines to optimize reproducibility.

Explore docker and windows subsystem for linux, virtual machines, and other container options to enable reproducible ml deployments and compare interoperability across windows, unix, and mac os.

Explore TensorFlow Extended and Airflow integrations to build, validate, and deploy ML pipelines for taxi data, covering data validation, feature engineering, training, and production deployment.

Perform batch prediction with PyTorch using a two-worker Dask cluster and lazy execution, then apply transfer learning for healthcare with asynchronous predictions and softmax probabilities.

Explore real-time streaming with Apache Spark MLlib, building pipelines with transformers and estimators to train, transform, and evaluate models on streaming data.

Explore shadow models for offline testing and retraining via AWS sage maker, integrating model monitor and data capture to track production performance across asynchronous and synchronous inference.

Explore embedding ml models on edge devices using SageMaker edge packaging, Greengrass, and Edge Manager to deploy, monitor, and infer on IoT devices for industrial safety use cases.

Deploy a PyTorch model via a Flask REST API, with image transforms and mapping predictions to image net labels, loading the model once for efficient post requests and SageMaker integration.

Demonstrates Streamlit for building intuitive web interfaces for ML models and data scripts, featuring a BERT keyword extractor and a YOLO v3 object detector with interactive controls.

Explore Coop Flow theory and compare Kubernetes-based machine learning pipelines with Airflow, Luigi, Argo, and more. See YAML-defined pipelines, training, deployment, and visualization like confusion matrices and ROC curves.

Explore production quality ml libraries—Hugging Face, Torch Rack, and Spark MLP for health care—and examine deployment, inference optimization, cpu and gpu acceleration, Onyx and rust tokenizers for low latency.

Review key ml ops concepts from section four and the course, including batch prediction, embedded models, and browser-based model interactions, while emphasizing production libraries and deployment considerations.