iiinova — where your digital self keeps evolving

Manuel Chevalier @manuelchevalier.bsky.social · 07/10/2026

I’ve fallen back on my favourite procrastination exercise: making maps. There’s something fun about making my own instead of handing the job to an app that decides how my data should look. My latest notebook entry: manuelchevalier.com/running-mapp... 📓

011

Manuel Chevalier @manuelchevalier.bsky.social · 05/10/2026

3/3 "We will provide data-analysis training" is a promise. A structured, network-wide curriculum is a credible plan. I can design it with you. DM me or email manuel.chevalier@datasharpacademy.com.

011

Manuel Chevalier @manuelchevalier.bsky.social · 05/10/2026

2/3 The goal is not to turn them into software developers. It is to make them independent enough to understand their work, evaluate results and collaborate. These technical skills are also transferable skills that improve employability.

111

Manuel Chevalier @manuelchevalier.bsky.social · 05/10/2026

1/3 Your MSCA Doctoral Network should not leave every candidate to invent their own way of working with data. Give them shared foundations in R and Python from the beginning.

112

Manuel Chevalier @manuelchevalier.bsky.social · 29/09/2026

3/3 I can help design and deliver that programme around your network's research and candidates. The initial discussion carries no commitment. DM me or email manuel.chevalier@datasharpacademy.com.

011

Manuel Chevalier @manuelchevalier.bsky.social · 29/09/2026

2/3 A credible, network-wide programme connects experimental design, data management, analysis, reproducibility, FAIR data and open sharing. It should complement local training, not duplicate it.

111

Manuel Chevalier @manuelchevalier.bsky.social · 29/09/2026

1/3 Writing an MSCA Doctoral Network proposal? Build data training in from the start, not after data collection. Once the data exist, some important decisions cannot be undone.

111

Manuel Chevalier @manuelchevalier.bsky.social · 23/09/2026

Is it permanent for 3 years, and then it ends? Sounds a lot like the old fixed-term contracts. 😅

010

Manuel Chevalier @manuelchevalier.bsky.social · 23/09/2026

Today I received this on a mailing list: "We are expanding our team at the Department of XXX in XXX and are opening a semi-permanent full-time Research Scientist position starting in 2027" I must be getting dumb, but could someone explain the concept of a semi-permanent job?

110

Manuel Chevalier @manuelchevalier.bsky.social · 21/09/2026

My challenge is to make them come to these conclusions without explicitly telling them, "you can't use it for that". Nobody wants to be told what to do, especially when the results often look good. Any suggestions?

022

Manuel Chevalier @manuelchevalier.bsky.social · 21/09/2026

As usual, I don't want to go into the details of how the models are built, but I'd rather make them think about the tool itself, what it does and doesn't do well, and where they can use it.

121

Manuel Chevalier @manuelchevalier.bsky.social · 21/09/2026

In a few weeks, I need to be teaching some AI to master's students for 6 hours Easy on paper, but I find myself struggling to find a good starting point. What can I teach people who, in all likelihood, are already familiar with the tool?

121

Manuel Chevalier @manuelchevalier.bsky.social · 18/09/2026

This week, I was told that I made statistics fun. Job done. 🫳 🎤

061

Manuel Chevalier @manuelchevalier.bsky.social · 16/09/2026

For me: keep training, collect more comparable efforts, and revisit the forecast. For you: run more/design new experiments, collect additional/complementary data, and formally test your hypothesis with a robust dataset. My latest notebook entry: manuelchevalier.com/running-five... 📓

manuelchevalier.com

The five miles I never ran

010

Manuel Chevalier @manuelchevalier.bsky.social · 16/09/2026

There is a parallel with preliminary research here. An early model can help us formulate a hypothesis and decide what evidence to collect next. And it brings a useful question : what does this analysis justify doing next?

manuelchevalier.com

The five miles I never ran

110

Manuel Chevalier @manuelchevalier.bsky.social · 16/09/2026

Quite a range, but this model still has value. Whether I will make the predicted time is not the most important. The model acts as a compass. My summer practice shows that I still have it. These fast paces I strive for are still achievable. Not there yet, but potentially there.

manuelchevalier.com

The five miles I never ran

110

Manuel Chevalier @manuelchevalier.bsky.social · 16/09/2026

In my latest chapter, I predict my 5-mile running time at the end of November. But I didn't run a single 5-mile at race pace in my calibration data. Net result: The predicted pace feels achievable (~4'45). But the uncertainties range from running slower than I do now to smashing the world record.

manuelchevalier.com

The five miles I never ran

110

Manuel Chevalier @manuelchevalier.bsky.social · 16/09/2026

In stats, I've always considered that most things are fair game. The good, the bad, and the ugly... ... as long as the assumptions and uncertainties are properly discussed. It is okay to make predictions with a few data points. But results should be discussed accordingly.

manuelchevalier.com

The five miles I never ran

131

Manuel Chevalier @manuelchevalier.bsky.social · 15/09/2026

I strongly believe this is the right approach for people who need stats for their work but have no stake in them. But... Will students be receptive? Or will they think I'm a fraud? I fear the Zoom death valley of black squares. ⬛️ ▪️ ◾️ 🔲 ◾️ ⬛️ 👨🏻‍🏫 🔲

010

Manuel Chevalier @manuelchevalier.bsky.social · 15/09/2026

First day jitters. Tomorrow I start teaching a new stats class. I know I know my s** t, and yet I am stressed. For this class, I decided to go with a completely different approach. Skipping all the maths and equations to focus only on what things mean. And how to use them.

271

Manuel Chevalier @manuelchevalier.bsky.social · 01/09/2026

It is a useful reminder that data science is rarely about complex models. Data Science is about understanding the problem well enough to put words to the story the data tell. If you know some properties of your data, don't restrain yourself. Use them first, and try to improve from there.

010

Manuel Chevalier @manuelchevalier.bsky.social · 01/09/2026

Sometimes, a few simple binary rules are enough. As a bonus, the results are remarkably easy to explain. A perfect score will rarely be achievable; today's results are just a sweet bonus. But even when simple rules sacrifice a little accuracy, the gain in interpretability often makes up for it.

110

Manuel Chevalier @manuelchevalier.bsky.social · 01/09/2026

I translated what I knew about the data and my running programme into a handful of explicit classification rules to see what would happen It turned out I didn’t need anything more. I didn’t even need to “model” the data. With these, every lap in the structured workouts fell into the right category.

110

Manuel Chevalier @manuelchevalier.bsky.social · 01/09/2026

Common sense before technique I had a classification problem for my running data. As I thought about which supervised-learning algorithm to use, I took a U-turn and tried the simplest, non-fancy solution first. --- My latest notebook entry: manuelchevalier.com/running-extr... 📓

manuelchevalier.com

Extracting the signal from the background

121

Manuel Chevalier @manuelchevalier.bsky.social · 26/08/2026

In this week's chapter, I describe my thought process in creating secondary variables from my rich running database. Dirty analyses were promised. Yet I hadn't delivered. After a series of fairly clean chapters on data engineering, here is my first proper excursion into exploratory data analysis.

010

Manuel Chevalier @manuelchevalier.bsky.social · 26/08/2026

You cannot create information that was never collected. But if the raw ingredients are there, you can often build an indicator that addresses your problem much more directly. That is where solid data-wrangling skills really pay off. (Note: Learn that before wanting to use any fancy model!)

110

Manuel Chevalier @manuelchevalier.bsky.social · 26/08/2026

Finding the right variable that addresses your question is difficult. And quite often, what you actually want isn't even there. At least not directly. I often notice how many students tend to stick to what they have. They forget they can engineer data columns. manuelchevalier.com/running-find...

manuelchevalier.com

Finding the right variable

111

Manuel Chevalier @manuelchevalier.bsky.social · 25/08/2026

What I want is for my students to be able to decide whether a technique makes sense for their question and their data. That means lots of guided practice, and even some role-play. It should be fun. I’ll teach this course for the first time in mid-September. Wish me luck.

010

Manuel Chevalier @manuelchevalier.bsky.social · 25/08/2026

For many people who use statistics, equations are another language to learn. But what if they didn’t need to start there? Who cares whether you can read the equation if you understand what the technique does, when it applies, and how to interpret it? Tools can handle the calculations for you.

110

Manuel Chevalier @manuelchevalier.bsky.social · 25/08/2026

Instead: - Understand what the technique does. - Assess whether it makes sense for your data. - Know why you may or may not decide to use it.

110

Manuel Chevalier @manuelchevalier.bsky.social · 25/08/2026

Over the past few weeks, I’ve been working on a course my younger self would have loved: 12 hours focused on the concepts behind statistical techniques and on deciding whether they make sense for your data. Not “hey, look at the equation; it says it all”.

111

Manuel Chevalier @manuelchevalier.bsky.social · 18/08/2026

My running notebook already has six chapters! As small and slightly irrelevant as it may be, I’m enjoying writing it. Just a little summer project, growing one chapter at a time ☀️📓 Chapters 7-10 are already brewing. I just need to run more to collect the data. buff.ly/IIObNCm

manuelchevalier.com

Running

Sixteen weeks of rebuilding a running routine while documenting every step of a complete data project, from data collection, interpretation, and race time predictions.

021

Manuel Chevalier @manuelchevalier.bsky.social · 12/08/2026

Their code feels less intimidating, errors become easier to trace, and moving or revisiting a project is no longer stressful. The world needs to know. That's why I adapted this chapter into the latest entry of my running notebook: 📓 A place for everything manuelchevalier.com/running-sett...

manuelchevalier.com

A place for everything

010

Manuel Chevalier @manuelchevalier.bsky.social · 12/08/2026

It may not sound like the most exciting topic, and I don't think it is immediately appreciated. I actually strongly suspect it isn't 😅 But it is consistently one of the parts students appreciate most at the end. Once they see a properly organised workspace, everything becomes clearer.

110

Manuel Chevalier @manuelchevalier.bsky.social · 12/08/2026

In my 𝗥 𝗙𝘂𝗻𝗱𝗮𝗺𝗲𝗻𝘁𝗮𝗹𝘀 𝗳𝗼𝗿 𝗗𝗮𝘁𝗮 𝗦𝗰𝗶𝗲𝗻𝗰𝗲 workshop, I dedicate an entire chapter to organising projects and scripts: separating inputs from outputs, creating reliable paths, naming files clearly, and writing scripts that can be rerun from a fresh R session.

110

Manuel Chevalier @manuelchevalier.bsky.social · 12/08/2026

Writing R code is rarely the hardest part of data analysis. Structure is. That's what makes things click ... or crash. The world needs to know. That's why I adapted this chapter into the latest entry of my running notebook: 📓 A place for everything manuelchevalier.com/running-sett...

manuelchevalier.com

A place for everything

122

Manuel Chevalier @manuelchevalier.bsky.social · 10/08/2026

All of this in a single pipeline and without my guidance. We had just finished our introduction to ggplot, and I had asked them to create a quick visual to practice ggplot's syntax. And they (successfully) went the extra mile. I'm still amazed they could achieve this in just four days 🤩

110

Manuel Chevalier @manuelchevalier.bsky.social · 10/08/2026

- They loaded their own dataset. - Processed it to engineer the columns needed for the plot. - Wrote the entire ggplot code. - And then fine-tuned the labels, colours, and overall style. - They exported the image to a different folder, demonstrating their understanding of file paths.

110

Manuel Chevalier @manuelchevalier.bsky.social · 10/08/2026

Four days prior, my student knew nothing about coding in R or any other language. A proper rookie. Yet, they managed to internalise everything I had taught so far, combining different pieces of the puzzle with their own data. - They wrote a very well-structured and reproducible script

110

Manuel Chevalier @manuelchevalier.bsky.social · 10/08/2026

Last week was the 5th online edition of my 𝗥 𝗙𝘂𝗻𝗱𝗮𝗺𝗲𝗻𝘁𝗮𝗹𝘀 𝗳𝗼𝗿 𝗗𝗮𝘁𝗮 𝗦𝗰𝗶𝗲𝗻𝗰𝗲 workshop. And I experienced such a proud teacher moment when one participant showed me this plot at the end of the fourth day. It might not look like much, but it truly is. I was really excited for them. Let me explain why.

121

Manuel Chevalier @manuelchevalier.bsky.social · 04/08/2026

The body is not meant to stay idle, neither sitting nor standing. You're just more used to sitting, so it hurts less (or less visibly)

010

Manuel Chevalier @manuelchevalier.bsky.social · 04/08/2026

So I always recommend building a solid understanding of the fundamentals instead of delegating it to AI. Even if it takes you much more time today. It will take you far. That's what I'm teaching this week. Mindset and Fundamentals first.

010

Manuel Chevalier @manuelchevalier.bsky.social · 04/08/2026

Many tools let you bypass technical bottlenecks and focus on the core task of making sense of your data. So the big challenge is still here, but at least you have fewer distractions. 😅 When you understand how things work, it gets easier to integrate new tools into your workflows.

110

Manuel Chevalier @manuelchevalier.bsky.social · 04/08/2026

In the past months, I've thought a lot about data science, its current ecosystem, and how much more efficient learning has become in recent years. (Without even talking about AI doing your mundane tasks.) manuelchevalier.com

manuelchevalier.com

Enter the Mind of a Data Scientist

My vision of Data Science. Follow complete investigations from idea to conclusion, where questions matter more than tools and thinking matters more than technique.

111

Manuel Chevalier @manuelchevalier.bsky.social · 30/07/2026

Among other things, we will cover: - Data types and structures - Loading your data in R - Organising your workflow for reproducibility - Master pipelines to connect your work - Data wrangling (cleaning data, aggregating fields, merging datasets) - Data visualisation with ggplot2

010

Manuel Chevalier @manuelchevalier.bsky.social · 30/07/2026

🚨 Last days before the workshop. 50% discount if you contact me here or by email. Take control of your data journey. Today. 📆 August 3-7, 2026 📍 Live & interactive, online 🎟️ datasharpacademy.com/workshop-rfu...

111