Konstantin Grotov

We can finally share the pre-print for our @msrconf Industry paper "Optimizing Duplicate Size Thresholds in IDEs": areyde.com/files/jupyter-…! Congratulations to @smtitov, Alexandr, @areyde, and @timofeybryksin, we are looking forward to presenting it in Australia! A thread.

RL training and coding-agent experiments not scaling locally? IdeGYM fixes that – and it's now open source. jb.gg/rsrch-idegym

Liner is partnering with @spoticlr at #ICLR2026 — supporting Best Paper and Travel Awards for LLM research. And to celebrate, we're giving away: ✈️ Round-trip flights + hotel to #ICML2026 in Seoul 🎁 $300 Liner Credits Follow @search_liner + repost to enter by 4/27. Liner is built for research workflows. Find papers, verify sources, and write with citations in one place. See you in 🇧🇷 and 🇰🇷! @iclr_conf @icmlconf

JetBrains Research democratizes environment setup with PIPer PIPer enables on-device sized models (Qwen3-8B) to match the performance of GPT-4o on environment setup tasks, achieving a 9x improvement over its base model at 25x lower cost.

However, the journey doesn’t stop here. We received insightful feedback on how to make the AI agent even more intuitive and user-friendly. A huge thank you to my coauthors Artem Borzilov, Maksim Krivobok, @timofeybryksin, Yaroslav Zharov 👏

In this work, we integrated an AI agent into JetBrains Datalore notebooks to simplify error resolution. A user study showed that participants found our AI-powered system significantly more effective for resolving errors compared to the non-agentic approach. 🤓

I'm happy to share that our paper, "Debug Smarter, Not Harder: AI Agents for Error Resolution in Computational Notebooks" has been accepted for oral presentation at @emnlpmeeting EMNLP’24 System Demonstrations! 🎉 🥳 📜 Paper: arxiv.org/abs/2410.14393

Models' contexts are getting so big that they can (and often should!) include an entire repository, while we are still evaluating them on methods and files. That's why we created Long Code Arena. Pre-print: arxiv.org/abs/2406.11612 Datasets: huggingface.co/spaces/JetBrai… Details in🧵!

Are you a data scientist? Can you spare 15 minutes for research on collaboration in data science? Participate in our survey tinyurl.com/2td2hxjy You can win a $50 voucher (raffle). Thanks for your help! #Python #R #DataScience @smtitov @sback_ @kgrotov @dhivyabharathi_

Just a reminder that the deadline for the IDE workshop, co-located with @ICSEconf 2024, is in two weeks (December, 7th). Anything related to IDEs, 4+1 pages short papers, 1 or 2+1 pages position papers. Let's build the coolest community together! All info: ide-workshop.github.io

You can find the details of the analysis in the paper and the supplementary materials. We hope that this rather simple pipeline can help other researchers compare the distributions of clones for other related languages, for example, JVM-based languages.

We found that Jupyter clones have more clones of larger size, and so do detect the same percentage of them, the threshold needs to be higher. This can be seen on the probability density function or on the quantile-quantile plot.

We can finally share the pre-print for our @msrconf Industry paper "Optimizing Duplicate Size Thresholds in IDEs": areyde.com/files/jupyter-…! Congratulations to @smtitov, Alexandr, @areyde, and @timofeybryksin, we are looking forward to presenting it in Australia! A thread.

Hey there! Our (very) short paper "Optimizing Duplicate Size Thresholds in IDEs" has been accepted to @msrconf Industry Track! This is actually the 5th MSR in a row for the ML4SE Lab! Thanks a lot to @smtitov , Alexandr, @areyde , and @timofeybryksin ! Pre-print coming soon.

Wow! My first academic paper in SE thematic and already with an award. We will develop the theme of notebooks further!

@thattommyhall Quickly looked through the dataset: close, but no!

@TheZachMueller Although that notebooks dataset was collected quite a long time ago (autumn 2020), but it contains approximately 2000 such notebooks. Stay tuned, we will soon post a new convenient way to study the dataset.

BTW @kgrotov I haven't read this yet, but did you happen to examine any nbdev repos during the study?

Oooo I'll be reading this one with great interest. Love this type of analysis

In terms of structure, Jupyter notebooks are generally less complex but more entangled, and also employ user-defined functions differently. In terms of style, notebooks have more stylistics issues, but some of them are not applicable to Jupyter code and need to be updated.

The main contributions: (1) a large dataset of ~850,000 licensed Jupyter notebooks, (2) a library for processing, analyzing, and storing notebooks, and (3) a large-scale analysis of the differences between notebooks and regular Python code in terms of structure and style.

Our paper titled "A Large-Scale Comparison of Python Code in Jupyter Notebooks and Scripts" has been accepted to the main track of @msrconf 2022!🥳🥳🥳Congratulations to @smtitov, @areyde, and @timofeybryksin! The pre-print is available on arXiv: arxiv.org/abs/2203.16718