MLJAR, Inc.

📖 𝐉𝐞𝐯 𝐝𝐨𝐞𝐬𝐧’𝐭 𝐰𝐫𝐢𝐭𝐞. 𝐈𝐭 𝐝𝐞𝐜𝐢𝐝𝐞𝐬. In this article, Piotr Płoński test Jev in Python for classification, scoring, and agent routing — and compare it with GPT-5.4 nano. He also look at an important question: > If Jev can’t generate arbitrary text, does that really mean it doesn’t hallucinate? Read the full article on our blog: 👉 https://lnkd.in/diFi2_Qg

The original iPhone cost $499 in 2007. The iPhone 17 costs $799. That looks like a 60% increase. Adjusted for inflation, it isn't. In August 2026 dollars: → Original iPhone: $806 → iPhone 17: $831 Eighteen years, and the standard iPhone costs the same in real terms. Since 2020 it has actually been getting cheaper, because Apple kept the price at $799 while prices everywhere else rose. The Pro line fell even further: $1,365 for the iPhone X, $1,067 for the 16 Pro. Then 2026 turned it around, with the 18 Pro at $1,199. So where does "iPhones got expensive" come from? The lineup. In 2007 there was one iPhone. Today there are models above and below it, up to $1,999 for the foldable Duo. The expensive iPhone is new, not the regular one. I adjusted every launch price with CPI-U from the BLS, fitted the trends in Python, and turned the notebook into an interactive dashboard with Mercury, so you can check any of it yourself. Full analysis and the dashboard in the comments 👇 #Python #DataAnalysis #pandas #DataVisualization

📚 A recent paper in the 𝐈𝐧𝐭𝐞𝐫𝐧𝐚𝐭𝐢𝐨𝐧𝐚𝐥 𝐉𝐨𝐮𝐫𝐧𝐚𝐥 𝐨𝐟 𝐅𝐚𝐭𝐢𝐠𝐮𝐞 shows how automated feature engineering can uncover useful patterns, while engineering knowledge is still needed to decide which ones actually make sense. paper: "𝘋𝘪𝘴𝘤𝘰𝘷𝘦𝘳𝘺 𝘰𝘧 𝘧𝘢𝘵𝘪𝘨𝘶𝘦 𝘴𝘵𝘳𝘦𝘯𝘨𝘵𝘩 𝘮𝘰𝘥𝘦𝘭𝘴 𝘷𝘪𝘢 𝘧𝘦𝘢𝘵𝘶𝘳𝘦 𝘦𝘯𝘨𝘪𝘯𝘦𝘦𝘳𝘪𝘯𝘨 𝘢𝘯𝘥 𝘢𝘶𝘵𝘰𝘮𝘢𝘵𝘦𝘥 𝘦𝘟𝘱𝘭𝘢𝘪𝘯𝘢𝘣𝘭𝘦 𝘮𝘢𝘤𝘩𝘪𝘯𝘦 𝘭𝘦𝘢𝘳𝘯𝘪𝘯𝘨 𝘢𝘱𝘱𝘭𝘪𝘦𝘥 𝘵𝘰 𝘵𝘩𝘦 𝘸𝘦𝘭𝘥𝘦𝘥 𝘵𝘳𝘢𝘯𝘴𝘷𝘦𝘳𝘴𝘦 𝘴𝘵𝘪𝘧𝘧𝘦𝘯𝘦𝘳" The authors ( Prof. Dr. Michael Kraus, Dr.-Ing. Helen Bartsch) studied fatigue strength prediction for welded transverse stiffeners and combined automated machine learning, feature engineering, and explainability. The authors used 𝗚𝗼𝗹𝗱𝗲𝗻 𝗙𝗲𝗮𝘁𝘂𝗿𝗲𝘀 from MLJAR AutoML as part of their feature engineering process. ⭐𝗚𝗼𝗹𝗱𝗲𝗻 𝗙𝗲𝗮𝘁𝘂𝗿𝗲𝘀 ⭐ 𝘢𝘶𝘵𝘰𝘮𝘢𝘵𝘪𝘤𝘢𝘭𝘭𝘺 𝘤𝘳𝘦𝘢𝘵𝘦 𝘯𝘦𝘸 𝘷𝘢𝘳𝘪𝘢𝘣𝘭𝘦𝘴 𝘧𝘳𝘰𝘮 𝘦𝘹𝘪𝘴𝘵𝘪𝘯𝘨 𝘰𝘯𝘦𝘴 𝘶𝘴𝘪𝘯𝘨 𝘰𝘱𝘦𝘳𝘢𝘵𝘪𝘰𝘯𝘴 𝘴𝘶𝘤𝘩 𝘢𝘴 𝘢𝘥𝘥𝘪𝘵𝘪𝘰𝘯, 𝘴𝘶𝘣𝘵𝘳𝘢𝘤𝘵𝘪𝘰𝘯, 𝘮𝘶𝘭𝘵𝘪𝘱𝘭𝘪𝘤𝘢𝘵𝘪𝘰𝘯, 𝘢𝘯𝘥 𝘥𝘪𝘷𝘪𝘴𝘪𝘰𝘯. 𝘛𝘩𝘦 𝘨𝘰𝘢𝘭 𝘪𝘴 𝘴𝘪𝘮𝘱𝘭𝘦: 𝘥𝘪𝘴𝘤𝘰𝘷𝘦𝘳 𝘳𝘦𝘭𝘢𝘵𝘪𝘰𝘯𝘴𝘩𝘪𝘱𝘴 𝘵𝘩𝘢𝘵 𝘮𝘢𝘺 𝘣𝘦 𝘥𝘪𝘧𝘧𝘪𝘤𝘶𝘭𝘵 𝘵𝘰 𝘪𝘥𝘦𝘯𝘵𝘪𝘧𝘺 𝘮𝘢𝘯𝘶𝘢𝘭𝘭𝘺. In the experiments, some of these automatically generated features improved cross-validation performance. But there was an important catch 🔬 . Some of the features had little or no physical meaning 🤷♂️ . A mathematical transformation can be useful for prediction while still being difficult to justify from an engineering point of view. The authors ultimately decided not to use those features in their final models. I think this is a great example of what practical AutoML should look like. ✅ AutoML can search a much larger space than a human can reasonably explore manually. It can suggest models, transformations, and relationships worth investigating. ✅ 𝗕𝘂𝘁 𝗱𝗼𝗺𝗮𝗶𝗻 𝗸𝗻𝗼𝘄𝗹𝗲𝗱𝗴𝗲 𝘀𝘁𝗶𝗹𝗹 𝗱𝗲𝗰𝗶𝗱𝗲𝘀 𝘄𝗵𝗲𝘁𝗵𝗲𝗿 𝘁𝗵𝗼𝘀𝗲 𝗿𝗲𝗹𝗮𝘁𝗶𝗼𝗻𝘀𝗵𝗶𝗽𝘀 𝗺𝗮𝗸𝗲 𝘀𝗲𝗻𝘀𝗲. For me, the interesting lesson is not that automation replaces the expert. It is that 𝗔𝘂𝘁𝗼𝗠𝗟 𝗰𝗮𝗻 𝗯𝗲𝗰𝗼𝗺𝗲 𝗮 𝗱𝗶𝘀𝗰𝗼𝘃𝗲𝗿𝘆 𝘁𝗼𝗼𝗹 𝗳𝗼𝗿 𝘁𝗵𝗲 𝗲𝘅𝗽𝗲𝗿𝘁 ⭐. link to the article 👉 https://lnkd.in/dXhQSsyR #AutoML #MachineLearning #ExplainableAI #Engineering #FeatureEngineering #DataScience #MLJAR

These are NOT GitHub commits. 🟩 📆 I turned more than 4 years of 𝐆𝐢𝐭𝐇𝐮𝐛 𝐢𝐧𝐜𝐢𝐝𝐞𝐧𝐭𝐬 into a GitHub-style activity calendar. Each square represents a day when a GitHub Status incident started. And once you see the data this way, patterns start jumping out immediately. The dataset contains: • 829 incident records, • 575 days with at least one incident, • March 2022 → September 2026. You can also filter the calendar by impact or GitHub component and switch between incident count and total incident duration. The interesting part? The whole interactive app is just a 𝐉𝐮𝐩𝐲𝐭𝐞𝐫 𝐍𝐨𝐭𝐞𝐛𝐨𝐨𝐤 + 𝐏𝐲𝐭𝐡𝐨𝐧 + 𝐩𝐚𝐧𝐝𝐚𝐬 + 𝐌𝐞𝐫𝐜𝐮𝐫𝐲. No separate frontend. Full analysis + code: 👉 https://lnkd.in/dn9HQ9nd

🤓 A research team just turned MLJAR AutoML into a reusable skill for AI agents. In the new AREX-Skill paper, researchers (Jianlyu Chen, Jiawei Liu, Zhicheng Dou, Yuyang Hu, Wenqing Wei, Chaozhuo Li, Zheng Liu, Hongjin Qian, Xiaolong Chen, Defu Lian, Qiwei Ye) from Beijing Academy of Artificial Intelligence(BAAI), University of Science and Technology of China, Renmin University of China, and Hong Kong Polytechnic University of Artificial Intelligence analyzed 1,000 machine learning GitHub repositories and converted them into more than 5,300 operational skills for AI agents. 𝐦𝐥𝐣𝐚𝐫-𝐬𝐮𝐩𝐞𝐫𝐯𝐢𝐬𝐞𝐝 is one of them. They didn’t just summarize the README. They extracted practical workflows around: • AutoML training • preprocessing • reports and model interpretation • fairness • model persistence • Mercury app generation They also created smoke tests to check that these workflows actually work. So an AI agent can now learn not only that MLJAR exists, but also when and how to use it. I especially liked how clearly the illustrations explain the journey from repositories to reusable agent skills. Open-source libraries are becoming more than packages developers import. They are becoming executable knowledge for AI agents. Paper 👉 https://lnkd.in/dTAwdEfP MLJAR AutoML 🌟 https://lnkd.in/gWYrk5N 🌟 MERCURY 🌟 https://lnkd.in/eV7UuMSS 🌟

We’ve just added 3 new visualization widgets to Mercury 🟩 ActivityCalendar — GitHub-style activity heatmaps, 🔀 Sankey — visualize flows and relationships, 🔻 Funnel — see how quickly a dataset narrows after applying filters. We didn’t want to test them only on toy datasets, so we built real examples: • GitHub incidents visualized with a GitHub-style calendar; • 14 years of programming language mentions from Hacker News “Who is Hiring?”; • A job funnel for 𝐫𝐞𝐦𝐨𝐭𝐞 + 𝐏𝐲𝐭𝐡𝐨𝐧 + 𝐃𝐚𝐭𝐚/𝐌𝐋 + 𝐬𝐚𝐥𝐚𝐫𝐲 𝐢𝐧𝐟𝐨𝐫𝐦𝐚𝐭𝐢𝐨𝐧. The workflow stays simple: 𝐉𝐮𝐩𝐲𝐭𝐞𝐫 𝐍𝐨𝐭𝐞𝐛𝐨𝐨𝐤 → 𝐏𝐲𝐭𝐡𝐨𝐧/𝐩𝐚𝐧𝐝𝐚𝐬 → 𝐟𝐢𝐥𝐭𝐞𝐫𝐬 → 𝐚𝐧𝐚𝐥𝐲𝐬𝐢𝐬 → 𝐯𝐢𝐬𝐮𝐚𝐥𝐢𝐳𝐚𝐭𝐢𝐨𝐧 → 𝐢𝐧𝐭𝐞𝐫𝐚𝐜𝐭𝐢𝐯𝐞 𝐰𝐞𝐛 𝐚𝐩𝐩 No separate frontend needed. More about the new widgets and the examples you can find here 👉 https://lnkd.in/dmFrhyKp Source notebooks: https://lnkd.in/dqVpz6qb #Python #Jupyter #DataVisualization #DataScience #OpenSource #Mercury

📚 A recent paper in the 𝐈𝐧𝐭𝐞𝐫𝐧𝐚𝐭𝐢𝐨𝐧𝐚𝐥 𝐉𝐨𝐮𝐫𝐧𝐚𝐥 𝐨𝐟 𝐅𝐚𝐭𝐢𝐠𝐮𝐞 shows how automated feature engineering can uncover useful patterns, while engineering knowledge is still needed to decide which ones actually make sense. paper: "𝘋𝘪𝘴𝘤𝘰𝘷𝘦𝘳𝘺 𝘰𝘧 𝘧𝘢𝘵𝘪𝘨𝘶𝘦 𝘴𝘵𝘳𝘦𝘯𝘨𝘵𝘩 𝘮𝘰𝘥𝘦𝘭𝘴 𝘷𝘪𝘢 𝘧𝘦𝘢𝘵𝘶𝘳𝘦 𝘦𝘯𝘨𝘪𝘯𝘦𝘦𝘳𝘪𝘯𝘨 𝘢𝘯𝘥 𝘢𝘶𝘵𝘰𝘮𝘢𝘵𝘦𝘥 𝘦𝘟𝘱𝘭𝘢𝘪𝘯𝘢𝘣𝘭𝘦 𝘮𝘢𝘤𝘩𝘪𝘯𝘦 𝘭𝘦𝘢𝘳𝘯𝘪𝘯𝘨 𝘢𝘱𝘱𝘭𝘪𝘦𝘥 𝘵𝘰 𝘵𝘩𝘦 𝘸𝘦𝘭𝘥𝘦𝘥 𝘵𝘳𝘢𝘯𝘴𝘷𝘦𝘳𝘴𝘦 𝘴𝘵𝘪𝘧𝘧𝘦𝘯𝘦𝘳" The authors ( Prof. Dr. Michael Kraus, Dr.-Ing. Helen Bartsch) studied fatigue strength prediction for welded transverse stiffeners and combined automated machine learning, feature engineering, and explainability. The authors used 𝗚𝗼𝗹𝗱𝗲𝗻 𝗙𝗲𝗮𝘁𝘂𝗿𝗲𝘀 from MLJAR AutoML as part of their feature engineering process. ⭐𝗚𝗼𝗹𝗱𝗲𝗻 𝗙𝗲𝗮𝘁𝘂𝗿𝗲𝘀 ⭐ 𝘢𝘶𝘵𝘰𝘮𝘢𝘵𝘪𝘤𝘢𝘭𝘭𝘺 𝘤𝘳𝘦𝘢𝘵𝘦 𝘯𝘦𝘸 𝘷𝘢𝘳𝘪𝘢𝘣𝘭𝘦𝘴 𝘧𝘳𝘰𝘮 𝘦𝘹𝘪𝘴𝘵𝘪𝘯𝘨 𝘰𝘯𝘦𝘴 𝘶𝘴𝘪𝘯𝘨 𝘰𝘱𝘦𝘳𝘢𝘵𝘪𝘰𝘯𝘴 𝘴𝘶𝘤𝘩 𝘢𝘴 𝘢𝘥𝘥𝘪𝘵𝘪𝘰𝘯, 𝘴𝘶𝘣𝘵𝘳𝘢𝘤𝘵𝘪𝘰𝘯, 𝘮𝘶𝘭𝘵𝘪𝘱𝘭𝘪𝘤𝘢𝘵𝘪𝘰𝘯, 𝘢𝘯𝘥 𝘥𝘪𝘷𝘪𝘴𝘪𝘰𝘯. 𝘛𝘩𝘦 𝘨𝘰𝘢𝘭 𝘪𝘴 𝘴𝘪𝘮𝘱𝘭𝘦: 𝘥𝘪𝘴𝘤𝘰𝘷𝘦𝘳 𝘳𝘦𝘭𝘢𝘵𝘪𝘰𝘯𝘴𝘩𝘪𝘱𝘴 𝘵𝘩𝘢𝘵 𝘮𝘢𝘺 𝘣𝘦 𝘥𝘪𝘧𝘧𝘪𝘤𝘶𝘭𝘵 𝘵𝘰 𝘪𝘥𝘦𝘯𝘵𝘪𝘧𝘺 𝘮𝘢𝘯𝘶𝘢𝘭𝘭𝘺. In the experiments, some of these automatically generated features improved cross-validation performance. But there was an important catch 🔬 . Some of the features had little or no physical meaning 🤷♂️ . A mathematical transformation can be useful for prediction while still being difficult to justify from an engineering point of view. The authors ultimately decided not to use those features in their final models. I think this is a great example of what practical AutoML should look like. ✅ AutoML can search a much larger space than a human can reasonably explore manually. It can suggest models, transformations, and relationships worth investigating. ✅ 𝗕𝘂𝘁 𝗱𝗼𝗺𝗮𝗶𝗻 𝗸𝗻𝗼𝘄𝗹𝗲𝗱𝗴𝗲 𝘀𝘁𝗶𝗹𝗹 𝗱𝗲𝗰𝗶𝗱𝗲𝘀 𝘄𝗵𝗲𝘁𝗵𝗲𝗿 𝘁𝗵𝗼𝘀𝗲 𝗿𝗲𝗹𝗮𝘁𝗶𝗼𝗻𝘀𝗵𝗶𝗽𝘀 𝗺𝗮𝗸𝗲 𝘀𝗲𝗻𝘀𝗲. For me, the interesting lesson is not that automation replaces the expert. It is that 𝗔𝘂𝘁𝗼𝗠𝗟 𝗰𝗮𝗻 𝗯𝗲𝗰𝗼𝗺𝗲 𝗮 𝗱𝗶𝘀𝗰𝗼𝘃𝗲𝗿𝘆 𝘁𝗼𝗼𝗹 𝗳𝗼𝗿 𝘁𝗵𝗲 𝗲𝘅𝗽𝗲𝗿𝘁 ⭐. link to the article 👉 https://lnkd.in/dXhQSsyR #AutoML #MachineLearning #ExplainableAI #Engineering #FeatureEngineering #DataScience #MLJAR

What’s up guys! Review of MLJAR Studio by Florian Dedov on Neural Nine YT channel https://lnkd.in/dBk7PfZj

📖 A recent paper published in Artificial Intelligence explores how different missing-data imputation methods can affect group fairness metrics in machine learning models. The authors used MLJAR AutoML as part of their experimental workflow. I like this kind of use case because it shows AutoML being used not only to search for good predictive models, but also as part of research on more responsible and trustworthy machine learning. Paper: “Exploring the influence of missing data imputation in group fairness metrics” written by Pedro Henriques Abreu, Arthur Dantas Mangussi, Ricardo Cardoso Pereira, Miriam Seoane Santos, .Ana Carolina Lorena, Mykola Pechenizkiy 🙌 👏 👏 Great work! 👉 https://lnkd.in/d9Uhif6B #AutoML #MachineLearning #ResponsibleAI #Fairness #DataScience #MLJAR