Yoan Jacquemin

Guess what Airbnb use to provision infrastructure for their data scientists. Kubernetes ! https://lnkd.in/ewDGpxUu Digital infrastructure for research remains a fertile ground for innovation and I am grateful to many colleagues for getting me into this privileged position where I have the opportunity to build and lead a truly interdisciplinary, highly motivated, "crack team" that want to put Lancashire and South Cumbria Integrated Care Board (ICB) on the map in fair and equitable collaborations with more well-established centres. Trusted Research Environments (#TRE) should allow iterative, agile development of features/microservices without burning the bank. Slick developer experience including the ability to train new research software engineers should be first class features. Converging too soon on too few a number of solutions/frameworks/accelerators risks stifling innovation. The more you think about all of this, the more obvious the answer becomes. Watch this space ;-)🚀🚀🚀 Christian Cole James Hetherington Simon Li George Svarovsky Philip Couch George Moulton Mike Harding Shaun Turner Stephen Dobson Saeed Umar

13

1 commentaire

For anyone building cloud lakehouses for healthcare, read the following analysis as well as keep abreast of the rapid innovation happening in this space. https://lnkd.in/emi4KDmC Edit: And this one from the MotherDuck / DuckDB team https://lnkd.in/e4WmdUZG (thanks Katy Sadowski ) I just tested running #OMOP analytical queries on DuckDB (small dataset, 200M rows, 8GB) using #Ibis in a Jupyter notebook running on Kubernetes. Aggregation queries run in very respectable times in spite of the fact that the data is stored on Microsoft Azure fileshare and mounted using CSI drivers rather than on local SSD. I can see some benchmarking work coming up if anyone in the OHDSI (Observational Health Data Sciences & Informatics) community wants to help. This will be a nice and simple study looking at the average size of databases on the IMI_EHDEN portal, creating multiple synthetic OMOP databases in #DuckDb, #Postgres and #DataBricks/#Snowflake of varying sizes covering the majority (5th to 95th percentiles maybe) and running #Achilles/#DQD or other benchmarks on them with varying compute sizes and comparing the results including performance, costs and limitations - e.g DuckDb does not allow concurrent writes from more than one process. This may also inform the design of future versions of the CDM and OHDSI tools. e.g. the CDM, cohort and results tables could sit on different database architectures (if they can't already). Frank DeFalco Ajit Londhe, MPH Clair Blacketer, MPH Paul Nagy Katy Sadowski Paul Nagy Patrick Ryan Craig Sachson - is someone doing this already? If not, happy to put together a brief proposal for a future community call. I will post this in the CDM working group as well.

19

2 commentaires

What is the most valuable quality for a biostatistician ?💡 Personnally, I would definitely say strategic thinking / anticipation ☝️ Most of our work is to anticipate the possible scenarios, the most plausible, but also the less plausible ones (because we are paid to deal with uncertainties, not certainties 🔎). Also, we need to manage data from patients who are doing everything well, but most importantly to manage well the others (😉 intercurrents events). We need to anticipate questions from regulators and the evolution of the regulatory framework (elsewhere it would be boring 🤷🏽‍♀️)… In brief, a very good reason to have your statistician ready since the idea of a study, to benefit entirely from his/her anticipation skills 🎯 Agree, disagree ? #biostatistics #clinicalresearch

239

21 commentaires

How to Lie with Statistics: Chapter 1 – Not All Samples Are a Vibe The Vibe: Your data is only as good as your sample. If your sample is sus (suspicious), your conclusions will be too. Red Flags to Watch For: Lazy Sampling 🙄: Grabbing data that’s easy to get, like asking only your Insta followers. That’s not “representative,” it’s just convenient. Ghosting Groups 👻: If you ignore certain people (like surveying only gym bros about fitness trends), you’re leaving out voices that could totally flip the script. Overloading One Side 🛠️: Stacking the deck by overrepresenting certain groups. For example, using a sample of techies to study “average screen time” is gonna make everyone else look chill. #medical #dental #physio #chapter01 #pharmD #pharmacy #LinkedInSeries #ProfessionalGrowth #FollowForMore #StatisticsForTheWin #ResearchTips #StatisticalThinking #LearningIsFun #JoinTheDiscussion

8

Many ML models are designed to predict patient survival (outcome Y) based on their condition data (X). In contrast, CCNets take a reverse approach: we simulate and identify the conditions (X) necessary to achieve a desired outcome, ensuring patient survival. For example, instead of applying a treatment (T) to try to improve patient survival, we can encode patient data (X) to learn latent variables (E). By starting with the desired outcome (Y), these latent variables (E) provide the additional information needed to create new conditions (X) that lead to the same desired outcome. Unlike inference methods that gradually approach the desired outcome, our generative causal model directly sets the target outcome of survival and uses the latent information (E) derived from patient data to determine the necessary conditions for achieving it. By following the data generation process, this generative model creates real-world scenarios and identifies the essential conditions that lead to the desired outcome for the patients. We provide the essential algorithms and supporting examples below. For further business opportunities and advanced design inquiries, please contact us: Git: https://lnkd.in/ge45Fn6V Algorithm Details: https://lnkd.in/g5_hA-bi

12

🌟 **Seth Godin's Blog Posts: A Kaggle dataset** 🌟 https://lnkd.in/eh6ewwpJ I’ve always been a fan of **Seth Godin**— his insights, storytelling, and unique way of looking at the world have inspired so many, myself included. His blog is a treasure trove of wisdom on marketing, leadership, and life! So, I decided to dive deeper into his work and create something I’m excited to share: a **Kaggle dataset** that contains all the blog posts Seth Godin has published on his current platform- https://seths.blog/ Whether you’re a data scientist, content marketer, or someone who loves Seth’s work, this dataset offers an exciting opportunity to explore patterns, analyze his writing style, or even find common themes across his posts. 🔍 **Why I Created This Dataset:** Seth's work has deeply influenced my thinking, especially in the realms of taking ownership and leading the way, irrespective of you having the position to do so. His words push boundaries, inspire action, and are incredibly insightful. By preparing this dataset, I wanted to give others a tool to explore and gain new insights into his content. 🚀 **What You Can Do With It:** - **Text Analysis**: Dive into the themes Seth writes about most frequently. - **Sentiment Analysis**: Uncover the emotional tone of his blog over time. - **Trend Spotting**: See how his focus has evolved in terms of topics and key phrases. - **Writing Style Study**: Understand the stylistic techniques that make his writing so impactful. Check it out, and I can’t wait to see how the data community engages with Seth’s incredible body of work. 💡 Interested? Visit my Kaggle profile to explore the dataset! #Kaggle #SethGodin #DataScience #ContentAnalysis #WritingStyle #TextMining #Leadership #Marketing

4

𝗔𝗡𝗡𝗢𝗨𝗡𝗖𝗘𝗠𝗘𝗡𝗧: EPISODE 9 (PART 1) IS HERE Time to understand the role of Prompt Engineers For the past 2 years, We’ve been in the revolution of AI, Being starstruck way back in late 2022 By the unbelievable outputs that could be produced By Generative AI tools like ChatGPT. Since then, it’s been the show of AI. Tools like ChatGPT, Gemini, MidJourney and many more Have made earlier seemingly complex tasks simpler. They have made re-imagine the possibilities. Huge shout out to content creators such as Ruben Hassid, Sami Sharaf, Zain Kahn (To name a few that I’ve followed & learned a ton from) And every other individual (ton of appreciation 👏) Who have explored these generative AI tools tirelessly And helped teach so many people How these tools could be used in various tasks & jobs But in all of this, Very rarely have I come across and got a chance to learn about the individuals who are behind the training of such advanced and sophisticated AI models. Those are 𝗣𝗥𝗢𝗠𝗣𝗧 𝗘𝗡𝗚𝗜𝗡𝗘𝗘𝗥𝗦. I’ve seen ads about “Prompt Engineer” being the next big role. But rarely have I come across articles or videos That could answer the following questions in detail - → Who are they ? → What are their primary responsibilities ? → What are their day-to-day tasks ? → What are the skills they possess ? → Why are they so important ? → What is the roadmap to become a Prompt Engineer ? To answer all these questions, today I bring to you the 9th episode of my podcast “𝗧𝗵𝗲 𝗦𝘁𝗮𝗿𝗹𝗶𝗴𝗵𝘁 𝗣𝗼𝗱𝗶𝘂𝗺” Where we’ll dive into the world of Prompt Engineers Through the lens and experience of Sourabh Acharya, An extremely dynamic individual, content creator And Prompt Engineer working at NVIDIA A big thanks to him for spending the time & effort needed to share these insights with the broad community and LinkedIn family. 𝗙𝗶𝗻𝗱 𝘁𝗵𝗲 𝗹𝗶𝗻𝗸𝘀 𝘁𝗼 𝘁𝗵𝗲 𝗲𝗽𝗶𝘀𝗼𝗱𝗲, 𝗻𝗼𝘄 𝗼𝗻 𝗬𝗼𝘂𝗧𝘂𝗯𝗲 𝗮𝗻𝗱 𝗦𝗽𝗼𝘁𝗶𝗳𝘆, 𝗶𝗻 𝘁𝗵𝗲 𝗰𝗼𝗺𝗺𝗲𝗻𝘁𝘀 𝘀𝗲𝗰𝘁𝗶𝗼𝗻 𝗯𝗲𝗹𝗼𝘄 P.S Stay tuned for Part 2, where we’ll dive into understanding how LLMs are trained and prompt engineering techniques that can guide the behavior of these models and produce desired output. Follow Varun Sagar Theegala for more such conversations with industry professionals to bring to you interesting insights and strategies from their experience. #ai #artificialintelligence #promptengineers #chatgpt #gemini #midjourney #promptengineering #careerinsights #thestarlightpodium

37

1 commentaire

Monitoring machine learning models for performance and data drift is crucial to ensuring accurate predictions. In this article, you will learn how to use the open-source NannyML library to estimate model performance and detect data drift for a diabetes classification model. https://lnkd.in/d3iuzYQh

4

This book presents the main chapters on data science and some very interesting ones on ordinal data manipulation!! Well, no need to mention the great "SAS"... #Unesp #Statistics #College #Library #Biostatistics #Medicine #SAS

2