November 19, 2019
Four short data-themed talks, sponsored by Ginkgo Bioworks (https://www.ginkgobioworks.com/), and hosted at DataRobot.
** Andrew Schonfeld, "D-Tale"
A new open source GUI tool for viewing pandas data structures, built with Python and React. https://github.com/man-group/dtale
** Zack Halloran, "Serving a PyTorch deep learning model on the web without a server"
How to set up a web accessible inference service using AWS Lambda and AWS API Gateway.
** Michael Punzalan, "Processing data outside DataFrames with custom types"
It's common to write data pipelines in SQL or objects with a pandas DataFrame-like interface. When it works it's great. But when table- or column-driven processing is an uncomfortable fit for the problem (i.e. iterating through records is a better fit), defining a type and endowing it with the methods you need (dunders and otherwise) can actually lead you to an implementation, make it easier to write correct and reusable code, and -- if you are inclined to write tests -- generate test cases useful for refactoring or reimplementation (in case you really want to have a SQL implementation, write a UDF, etc.).
** Abraham Polk, "Twitter sentiment analysis"
Using NLP and sentiment analysis to predict the sentiment of Twitter tweets based on the words in the tweets, using emojis to train the model. Demonstrates how even relatively simple models can have strong predictive power when combined with good machine learning libraries.
Meetup link: https://www.meetup.com/bostonpython/events/265373222/