The BAUG wrapped Q4 with presentations from David Cooperberg (Product Manager, Alteryx) and Joshua Marsh (BAUG co-leader). David gave a presentation on Generative AI -AiDIN and Alteryx while Joshua spoke on Machine Learning methodologies and Alteryx. Here is a recap from both speakers:
David:
In the presentation led by David, he introduced Aiden! Aiden is the AI engine that Alteryx is building that infuses the power of both generative AI, which is more recent and machine learning across the entire suite of products that make up the Alteryx cloud platform. David highlighted the inspiration drawn from recent generative AI trends, leading to the development of Aiden.
The innovative features encompass Workflow Summary, Open AI connectors, Magic documents, and AiDIN Copilot, showcasing accelerated workflow creation through generative AI within Alteryx Designer, Cloud, and Intelligence Suite. Additionally, AI Studio was unveiled as the backbone, offering users the capability to customize and seamlessly integrate generative AI models. David emphasized the Design Partner Program, an initiative inviting collaboration for refining these AI products (email GenAITeam@alteryx if your interested in joining). Here is a quick recap of the technologies David went over:
-
Workflow Summary: A tool that you can put on your designer canvas that when the workflow runs, we'll make a call out to chatGPT
-
Open AI Connectors: Integration tools enabling seamless interaction between Alteryx workflows and Open AI, facilitating tasks like text classification and sentiment analysis.
-
Magic Documents (within Auto Insights): A feature generating PowerPoint decks or emails summarizing dashboard information, enhancing business intelligence and reporting.
-
Aiden Copilot: An innovative AI-powered interface within Alteryx Designer, expediting workflow creation by providing intelligent recommendations and automating tool configurations.
Joshua:
In the presentation led by Josh, he gave an overview of machine learning methodologies using Alteryx's ML tools. Josh took us through a binary classification Use Case using NBA data that aimed to predict the result of a game's outcome (W or L). Josh noted that the first and most important step of the process is to define the problem's definition. After this operation he gave insight into the automated process of extracting, transforming, and loading the data prior to analysis. Josh emphasized that this process is usually the hardest to preform in production. After walking through the initial steps of curating the data Josh dove into Data exploration and Feature engineering. Both of these steps used visual analysis as an aid to understanding the data that was collected. The last step Josh overviewed was modeling the data using two classification models, Decision Tree and Logistic Regression. After both models were fit to the data, Josh walked through the model comparison tool that allowed users to see an overview of each model's performance and the underlying reason of selecting one over the other. Here is a quick recap of the methods Josh went over:
-
Problem Definition - Scope the problem and determine measures for success.
-
Data Logging and Loading (ETL) - Identify data, algorithms typically need all data to be numeric, fill missing values, remove duplicates, standardize.
-
Exploratory Data Analysis - Summarize and visualize the data to understand it (min, max, median, mean, quartiles, outliers, trends, etc.)
-
Feature Engineering - Create new input features from existing ones.
-
Model Selection and Training - Split the data into train, validation, and test sets. Select the proper model or models to train the dataset.
-
Model Testing - Inner loop testing and refining to produce better and better results.
