5 Key Takeaways from PyData Eindhoven 2022 that No-one Talks About
The data science team at Sia Partners attended PyData Eindhoven 2022, an annual gathering of data scientists, machine learning experts and other Python professionals. With an impressive line-up of companies there were plenty of use cases, learnings and tools to digest. That’s why we created a list of 5 key takeaways that tell you everything you need to know about the latest developments in the field.
?? A New Trend: Creating Value First, Building Fancy Models Second
In PyData’s keynote presentation, Marijn Markus explains how his projects helped people across the globe. The recurring theme in his presentation is that his solutions were not the most accurate, scalable or efficient. Instead, the projects he tackled were in areas that had no data solution in place at all. In this case, going from no solution to even a simple data solution gave an immediate, tangible benefit. This goes against most data scientists’ intuition of always implementing the model with the highest performance. Getting something simple in place allows you to iterate quickly and immediately reap benefits. A simple model is better than no model.
The attitude of solution first, method second is slowly making its way through the industry, with professionals realising that delivering business value is what matters most. This is especially true when delivering AI projects for good, as we do at Sia Partners, where the focus is on helping others and saving our planet, not building fancy models.
?? Schiphol Airport Paves the Way for Machine Learning-Powered Security Camera Systems
Especially interesting this year was Schiphol’s computer vision use case, where they demonstrated how even complex information, such as whether or not a plane is being refuelled, can still be extracted from video footage. By using real-time camera feeds and deep learning computer vision models, Schiphol was able to better understand the inefficiencies during the turnaround process and preemptively detect delays, thereby decreasing the monetary impact.?
Given the ubiquity of security cameras, this opens up a multitude of possibilities, such as analysing crowd data in public spaces or optimising logistic operations in real time. We are excited at the prospect of working together with Sia Partners’ clients to implement a next-gen machine learning solution using video data.
?? ASML Leverages Predictive Maintenance to Minimise Downtime
Of the many presentations at PyData Eindhoven, that of ASML stood out as the only one that actually linked their data science solution to a tangible business benefit, namely the amount of downtime saved by their models. Implementing a predictive maintenance pipeline to maximise system utilisation and reduce unexpected downtimes provided the business value that ASML’s clients were after.?
Most production processes have machines with many sensors, and so ASML’s approach could be applied in many different contexts. Our team knows from experience that building a predictive maintenance model should always be done with the monetary impact in mind, to convince as many stakeholders as possible of the project’s value.?
?? Just Eat Takeaway Uses Causal Inference To Optimise Pricing When Trials Are Impossible
A/B testing is a great way to discover the preferences of your customers, especially when it comes to pricing. That’s easier said than done for Just Eat Takeaway, a platform with mixed pricing autonomy where some restaurants are free to choose their own pricing. Instead, Just Eat Takeaway uses a method of causal inference and scenario generation to better understand the outcomes of certain business decisions. Rather than building a complicated machine learning model and then figuring out how it can answer business questions later, Just Eat Takeaway started with the business end in mind.?
Using causal inference and scenario generation, often in the context of Bayesian modelling, is an approach that our data team is all too familiar with. Unlike machine learning models that are fashionable in the data science industry, Bayesian models start with a hypothesis and deliver an explainable result. Especially useful when management needs a clear answer to their business questions.
?? Taipy: A New Challenger To Streamlit
We are all familiar with Streamlit, R Shiny and Python Dash, all great frameworks for quickly building a web app to deploy your models. During this conference we discovered a new challenger: Taipy. A framework focused on improving the areas where Streamlit lacks – design flexibility and layout control – while being just as fast and intuitive to use. After using this new tool at the Taipy booth, we were informed of a new release in January of 2023 aiming to bring some unique features to the package.
When building interactive dashboards for our clients we like to find a balance between scalability and ease of use. Taipy appears to find the right balance, with many frameworks falling short in either of the two categories. Ideally we would like to implement a solution that is reliable, flexible to the client’s needs and can scale easily to more users.
?? Our data science team is always looking for driven experts and young professionals to strengthen our team. If you like attending conferences, geeking out about ML and making an impact in large organisations using data, feel free to reach out on LinkedIn.
? We would love to hear about your experience at a data science conference. Which data science event should we attend next? Let us know in the comments!