Joined November 2007
- Tweets 53,255
- Following 1,259
- Followers 1,266
- Likes 30,607
1,038 Photos and videos
Kozo Nishida | 西田孝三 retweeted
🆕 Introducing the MDN MCP server!
Bring MDN's docs and browser compat data straight into your AI agent or IDE for accurate, up-to-date answers about the web platform.
No more outdated guesses from training data.
Learn more 👇
developer.mozilla.org/en-US/…
ALT A headline Introducing MDN MCP server above a terminal window titled Light-dark CSS function for images, showing a prompt asking how to use the light-dark() CSS function for images and which browsers support it, with a note that the mdn tool was called 4 times, followed by an answer explaining that light-dark() takes two image values and works anywhere an image is accepted. The MDN logo appears on the right.
19
50
2,091
Kozo Nishida | 西田孝三 retweeted
Choosing the wrong join can silently remove observations, duplicate rows, or introduce missing values into your data.
This is why understanding joins is such an important part of any data wrangling workflow.
The visualization below demonstrates some of the most commonly used joins in dplyr:
🔹 full_join() keeps all observations from both datasets and fills unmatched values with missing values (NA).
🔹 inner_join() keeps only observations with matching IDs in both datasets.
🔹 left_join() keeps all observations from the left dataset and adds matching values from the right dataset when available.
🔹 right_join() keeps all observations from the right dataset and adds matching values from the left dataset when available.
These joins are extremely common in practical data science projects and are essential for combining information from multiple data sources correctly. There are many additional types of joins available, but these are among the most frequently used in everyday data analysis.
I have just released a brand-new module in the Statistics Globe Hub covering this topic in detail, including practical examples in R.
The Statistics Globe Hub is my ongoing learning program focused on practical skills in statistics, data science, AI, and programming with R and Python.
More information about the Statistics Globe Hub: statisticsglobe.com/hub
#rstats #datascience #statistics #programming #tidyverse #dplyr #analytics #statisticsglobehub
7
31
702
Kozo Nishida | 西田孝三 retweeted
Jun 12
Can an AI agent run a real mass-spec pipeline?
We benchmarked our open-source pyOpenMS skill across 250 agent runs on 10 tasks.
A strong model went 96% → 100% with ~12x fewer API errors. A cheap model jumped 74% → 88% at half the cost.
Full study: k-dense.ai/blog/benchmarking…
1
2
12
626
Kozo Nishida | 西田孝三 retweeted
Jun 12
I spent 4 years writing pyOpenMS to wrangle mass-spec data, so I had to test this: can an AI agent actually run these pipelines? We did 250 agent runs over 10 real MS tasks, with vs without our open-source pyOpenMS skill. On a strong model it went 96% → 100%.
1
4
10
874
AMD Ryzen AI Halo. The ultimate local AI developer platform.
Pre-order now: bit.ly/4xv5PJS
⚡ Up to 128GB unified memory
⚡ Support for models up to 200B parameters
⚡ Windows & Linux support
⚡ Ready-to-run AI workflows out of the box
Build, prototype, and deploy locally without cloud constraints.
218
298
2,902
431,746
Jun 12
MERIT-ML: A Machine-Learning-Readiness Framework for Tabular Public Metabolomics Data | ChemRxiv chemrxiv.org/doi/full/10.264…
1
131
Python Monthly Topics「PrefectではじめるPythonワークフロー・フレームワーク」by @satoru_kadowaki 公開
gihyo.jp/article/2025/08/mon…
4
9
1,839
💡 【 COSCUP 2026 CFP Results Are Out! 📥 】
All review results have been sent to applicants' emails!
Due to limited time & venue capacity, we regret that we couldn't include all of the many excellent proposals. We sincerely thank everyone for bringing your tech practices to the community and driving the open source ecosystem forward. ✨
📢 【 Selected Speakers: Action Required 】
Please check your inbox (and spam folder 🫣) and click the confirmation link to complete your "Attendance Confirmation".
📌 Important Dates:
▪️ 06/23 (Tue) International speaker visa invitation deadline
▪️ 07/11 (Sat) Official schedule release
Regardless of how you participate, we can't wait to see you this summer!
1
1
7
401
Congrats to @GoogleDeepMind on the launch of DiffusionGemma.
The model generates 256 tokens in parallel per step, delivering 150 TPS on DGX Spark, and 1,000 TPS on a single H100.
We're supporting it from day one with:
• BF16 and NVFP4 checkpoints on @huggingface🤗
• Free GPU-accelerated endpoints on build.nvidia.com
• @vllm_project support with FP8 precision
Get started with DiffusionGemma on NVIDIA: nvda.ws/43ro19u
DiffusionGemma, our experimental open model released under an Apache 2.0 license, explores text diffusion, an exceptionally fast approach to text generation.
Here’s how DiffusionGemma accelerates development:
Faster token output: By shifting the bottleneck from memory bandwidth to raw compute, the model generates up to 4x faster token output on dedicated GPUs
Accessible hardware footprint: Activates just 3.8B parameters during inference, fitting comfortably within 24GB-VRAM high-end consumer GPUs when quantized
Novel workflows: Parallel token generation enables self-correction, making it ideal for code infilling, in-line editing, and non-linear structures
DiffusionGemma prioritizes speed over raw quality and accelerates best on compute-bound hardware (like @NVIDIAAI GPUs). Standard @GoogleGemma 4 remains recommended for production quality and memory-bound devices.
37
118
1,363
99,356