In this practical, hands-on training seminar participants will build their own cloud-based Natural Language Processing (NLP) applications from scratch. By the end of this course participants will understand cloud-based architecture, the principles of machine learning algorithms and their applications to NLP, and how to successfully build and implement NLP to suit each participant’s unique needs. Participants are guided through the entire process of building working applications, with someone at their side to resolve real-world problems such as configuration settings, library installation issues, and coding errors.

There are no formal prerequisite skills for this course, but a strong analytics mind and prior exposure to Python is helpful (for those interested, we can recommend short, 1- to 2-hour tutorials to bring a participant up to speed on Python fundamentals). As this course employs cloud-based infrastructure, participants need only bring with them a Windows/Linux/Mac laptop.

Day 1, morning session 1: Cloud-Based Development, part 1

Review student applications and data

AWS console

Security and user permissions

Manage EC2 instances & S3 buckets

Amazon SageMaker

Day 1, morning session 2: Cloud-Based Development Environment, part 2

Python, virtual environments, Jupyter notebooks

Version control and Git

Practice with Python, Jupyter, and git

Day 1, afternoon session 1: Sourcing and Managing Data

Sourcing and validating data

Managing training and test data

NLP data for sentiment

Creating training data: AWS training data tools

Day 1, afternoon session 2: Practice & Review

Practice with Python, Jupyter, git, and AWS training data tools

Shape training data

Day 2, morning session 1: Introduction to NLP

NLP applications

Preprocessing: stemming, tokenization

Machine learning algorithms

Initial NLP models

Day 2, morning session 2: Practice & Review

Implement stemming, tokenization with student data

Configure machine learning algorithms

Day 2, afternoon session 1: Transfer Learning and Training Models

Word embedding with ULMFiT

Transfer learning with fast.ai

Training NLP models

Model tuning and hyperparamters

Day 2, afternoon session 2: Batch Inference

Accessing model artifacts

Deploying models on large data

Output results (json, csv, Redshift)

Day 3, morning session 1: Practice & Review

Accessing model artifacts

Deploying models on large data

Output results (json, csv, Redshift)

Day 3, morning session 2: Restful Inference

Restful architecture

Lambda functions

Day 3, afternoon session 1: Practice & Review

Implement restful architecture with Lambda using student data

Day 3, afternoon session 2: Deployment

Docker instances

Practice using Docker with student applications

Day 4, morning session 1: Application Architecture

End-to-end application architecture including persisting and accessing model results

Performance optimization, AWS hosting, and GPUs

AWS economics

Day 4, morning session 2: Build Student Applications

Review specific topics as a class or small groups

Guided development and problem solving

1.1 development coaching

Day 4, afternoon sessions: Build Student Applications

Guided development and problem solving

Optional 1.1 development coaching

Ensure each student leaves with a functioning application