Doing a basic web deployment of a deep learning model is good way to prototype how your model will be used and to validate assumptions that you made during the training process. While working with Keras, I did simple deployments of several models and published articles describing my experience deploying Keras models with Facebook Messenger and Flask.
Since I have been spending a lot of time in the last year using the other major high-level deep learning framework, fastai, I wanted to get practical experience with simple deployments of fastai models. I decided to adapt the the approach I used…
In late 2020 I published an article on using GPT-3 to navigate the London Underground. This article described how I created a simple harness in Python that:
If you want to learn about deep learning you really can’t go wrong with the fastai framework. This framework is at the heart of a set of courses and is now the topic of a book written by the fastai leader Jeremy Howard.
Here are some of the benefits of fastai:
I have done a few simple experiments with GPT-3, including comparing its performance to a Rasa chatbot and using it to generate git commands. From this experience, and the dozens of applications of GPT-3 that others have published, I have come to appreciate GPT-3’s remarkable ability to solve a wide range of problems. I had not, however, seen GPT-3 tackle any spatial problems, so I asked myself, how well would GPT-3 be able to tackle a spatial navigation problem?
I thought that navigating a subway system would be a good test case for GPT-3’s spatial abilities because trips on a…
There are some basic skills that are easy if you learn when you are young but can be baffling if you have to master them later in life. For example, driving a car with a manual transmission can seem to be perversely difficult to somebody who learned to drive on an automatic. However, the challenge of “driving stick” pales in comparison to mastering the git command line interface if you didn’t grow up with git. In this article I’ll describe how I harnessed OpenAI’s GPT-3 to generate git commands from English language descriptions.
Git is a foundational technology and a…
In 1829, an event took place that unleashed a technological revolution. At the Rainhill Trials a group of steam locomotives squared off to determine which one could win a series of tests of speed, strength and reliability. The winning machine, Rocket, not only blew away its competition at the trials, it also set the direction for steam locomotive development for the following century.
Relational databases like Postgres include a set of tables that describe the tables in the database. This set of metadata tables is called the catalog and it contains a treasure trove of details about the database. I recently needed to write a program to automatically extract insights from database catalogs and this led me to write a simple Python module to connect to a Postgres database, get information from the catalog, and load that information into a Pandas dataframe for further processing in Python. This article describes the process I followed.
One of the key facts about Postgres is that…
I recently had the opportunity to deliver a hands-on workshop on training a Keras deep learning model. This workshop was a follow-on for a session I had done for a local meetup that reviewed the content in my upcoming book for Manning Publications, Deep Learning with Structured Data. After the introductory session there was appetite for a hands-on session where participants would be able to work through the process of training training one of the deep learning models featured in the book.
Ever since I had my first taste of deep learning I have been interested in applying it to structured, tabular data. I have written several articles on the subject and I am writing a book on Deep Learning with Structured Data for Manning Publications. It would be great to tackle problems with structured tabular data by harnessing deep learning’s flexibility and potential for reduced feature engineering.
The idea of using deep learning on tabular data is not without its critics. A consistent objection I have heard is that non-deep learning approaches, XGBoost in particular, are simpler to code, easier to…
In a previous article I described the process of deploying a deep learning model using Facebook Messenger and the Rasa chatbot framework. This approach provided a smooth user experience for the model in question (predicting delays on Toronto’s streetcar network), but it has a lot of moving parts: configuration of a Facebook app, training a Rasa chatbot, and the definition of custom actions for the chatbot in Python to apply the deep learning model to get a prediction of whether or not the streetcar trip the user wants to take will be delayed.
Data Science manager at Intact Insurance. Opinions expressed are my own.