A Journey into PyCon Italia 2023
PyCon is the largest annual gathering for the community using and developing the open-source Python programming language and it is held over the World.
At the last weekend of May ’23, “PyCon Italia” organized the PyCon event in Italy. Looking at some highlights, it started with two commercial talks: “GPT or how I learned to stop worrying and love the generative AI” by Tommaso Radicioni and “Beyond grid search: XGBoost and Optuna as the ultimate ML optimization combo” by William Arias.
Great explanation by Tommaso, he started with the evolution of Language Models from n-gram to GPT as Large Language Models. He went deeply showing how to build and improve a GPT model into the Hugging-Face platform, and in my opinion, this talk would be worth a workshop.
I was impressed by William’s presentation by the simplicity of how he talked about the Optuna optimization system applied to the XGBoost model. You just need to embed your model and hyperparameters into an objective function and everything is done!!!
One strength of this conference was the atmosphere, good food, red wine at every table for lunch and friendly people.
A weakness that I met was the organization of the rooms, with a crowd of people, and so many times seats weren’t enough.
Friday's day ended with a PyDrink with the opportunity to enjoy with other Pythonists discussing on the talks presented during the day.
Saturday was really special, with “Beyond Pandas: lightning fast-in memory data frames with Polars” by Alberto Danese. Surely the best talk that I’ve listened to and one of the best presentations than the others that I haven’t had the opportunity to listen to.
Polars is the new attractive tool to manage data frames faster than Pandas with a Pythonic language, so a really good alternative to it. Alberto explained some Polars key features and was able to explain the weaknesses and strengths between Polars, Pandas and others. I think it’s time to try this tool.
PyCon Italia organized also a mini walk tour in Florence for those who were interested to visit the beautiful city, and the day ended with a PyDinner (eating Fiorentina for fans) and a concert. Amazing Italian mood!!!
The event ended, for this edition, on Sunday with interesting talks to mention. First of all “ChatGPT and Minecraft’s diamonds playing agents: Reinforcement Learning in a nutshell with sheeprl” by Federico Belotti, he spoke about SheepRL, a Deep Reinforcement Learning framework built by Orobix on the top of OpenAI Gym. Going ahead with Orobix guys is mandatory to mention “Controlling the Chaos: Reproducible Deep Learning Experiments Using Lightning and Hydra” by Refik Can Malli. In this talk, Refik explained how to combine YAML files and open-source tools such as Hydra, Lightning framework, and Pytorch to make more efficient and reproducible deep learning projects. Another good talk that would be worth a workshop. Then surely the topic to pay attention to in the future is AI Safety and Stefania Delprete gave us just a little taste with the talk “Introduction to AI Safety. Let’s be ready before it’s to late!”
I want to mention my other favourite speech coming from Luca Baggi: “Is the great data frame showdown finally over? Enter: polars”. With this talk, Luca went deeply into Polars showing lazy and eager APIs in query applications going beyond the introduction held by Alberto Danese.
I was not attracted by the workshops and I hope in future meetings to see something able to catch up my attention.
You can enjoy the conference with youtube videos, and pick up inspirations for your job.
Written by Claudio G. Giancaterino
