Andy Rowlinson
@numberstorm.bsky.social
Data insights and creator of mplsoccer, a Python library for football viz.
This is looks super interesting. Lots of docs that explain to an LLM how to use the python data stack.
github.com/K-Dense-AI/c...
github.com/K-Dense-AI/c...
claude-scientific-skills/scientific-skills at main · K-Dense-AI/claude-scientific-skills
A set of ready to use scientific skills for Claude - K-Dense-AI/claude-scientific-skills
github.com
November 19, 2025 at 6:47 PM
This is looks super interesting. Lots of docs that explain to an LLM how to use the python data stack.
github.com/K-Dense-AI/c...
github.com/K-Dense-AI/c...
Making some nice progress on the NFL pitches. 🏈
November 2, 2025 at 5:39 PM
Making some nice progress on the NFL pitches. 🏈
I'm going to see if I can release something for 🏈. It should have all the usual viz methods you love in mplsoccer, plus customizable stuff like endzone colours etc.
October 26, 2025 at 9:15 PM
I'm going to see if I can release something for 🏈. It should have all the usual viz methods you love in mplsoccer, plus customizable stuff like endzone colours etc.
Browsing the @kaggle.com big data bowl, it looks like there aren't good libraries for plotting NFL data.
October 26, 2025 at 9:15 PM
Browsing the @kaggle.com big data bowl, it looks like there aren't good libraries for plotting NFL data.
You can also use the code if you want. I made the license MIT so it allows that kind of stuff.
October 11, 2025 at 6:39 AM
You can also use the code if you want. I made the license MIT so it allows that kind of stuff.
Of course, someone already did a plotly package a while back, I don't know how good it was as I never tried it.
October 11, 2025 at 6:39 AM
Of course, someone already did a plotly package a while back, I don't know how good it was as I never tried it.
💜 this might be just what I need to finish one of my projects.
October 8, 2025 at 3:57 AM
💜 this might be just what I need to finish one of my projects.
That looks great!
October 1, 2025 at 4:15 AM
That looks great!
They are used a lot for comparing models, e.g. bsky.app/profile/sung...
One thing I like about Nvidia is that they open-source all their work—models, code, and datasets. It makes sense, since their real business is selling shovels: the more people digging for gold, the better it is for them.
Nvidia's Jet-Nemotron
Nvidia's Jet-Nemotron
October 1, 2025 at 4:13 AM
They are used a lot for comparing models, e.g. bsky.app/profile/sung...
This is the kind of thing that I think it could help with sifting through raw text and cleaning it up
bsky.app/profile/jere...
bsky.app/profile/jere...
I like Claude Code for this kind of thing. I doubt I would have found and fixed 151 data cleaning issues in 13 minutes.
![⏺ Task(Analyze case name patterns)
⎿ Done (15 tool uses · 48.9k tokens · 12m 58s)
⏺ I've completed a comprehensive analysis of case names across all three CSV files (8,707 total cases). Here's what I found:
Key Findings
151 unique issues identified across these categories:
Major Issues:
1. United States typos (22 cases): "United State", "United Staes", "United Sates", "United Statse", etc.
2. State name typos (55 cases): Virgina, Tennesse, Georiga, Conneticut, Illnois, Minneosta, Pennsylvannia, etc.
3. Concatenated "of[State]" (16 cases): "ofMassachusetts", "ofVirginia", "ofPennsylvania"
4. "American" instead of "America" (4 cases): "United States of American"
5. "vs" instead of "v." (13 cases): Mostly Oregon and North Carolina
6. Whitespace issues (23 cases): Leading/trailing spaces
7. Multiple consecutive spaces (4 cases)
8. Placeholder values (13 cases): "Unknown", "Data not available", "W"
Your Current Code Coverage:
✅ Already handles: Some "United States" typos, "Ofmassachusetts", Tennessee variations❌ Missing: Most state typos,
whitespace issues, "vs" patterns, "United States of American", placeholders
I've provided a complete updated cleaning code block that fixes all 151 issues. The code is ordered properly (whitespace →
concatenation → spelling → separators → standardization) and includes all the fixes your current code has plus the new ones
needed for the two additional datasets.](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:yktvw5fpe6j4rqy4vio7l3w3/bafkreidct4memo7q2ntv3bjifokuzyyhkuawna4nv7ockcmn7mfmykeuzu@jpeg)
October 1, 2025 at 4:07 AM
This is the kind of thing that I think it could help with sifting through raw text and cleaning it up
bsky.app/profile/jere...
bsky.app/profile/jere...
That's probably why I find it less helpful right now because I made a plotting library and know most of the quirks of matplotlib. I'd really like to find a way of doing this 90% AI assisted type stuff for data science. Maybe we need some better prompts
September 30, 2025 at 3:26 AM
That's probably why I find it less helpful right now because I made a plotting library and know most of the quirks of matplotlib. I'd really like to find a way of doing this 90% AI assisted type stuff for data science. Maybe we need some better prompts
I have a library for doing this quite nicely in Python, it's a sport library but I may split it off one day.
mplsoccer.readthedocs.io/en/latest/ga...
mplsoccer.readthedocs.io/en/latest/ga...
September 30, 2025 at 3:14 AM
I have a library for doing this quite nicely in Python, it's a sport library but I may split it off one day.
mplsoccer.readthedocs.io/en/latest/ga...
mplsoccer.readthedocs.io/en/latest/ga...
Do you have different models you are trying compare? If you can get more than one metric, I would use a radar. I found this one for RAG from
www.deepset.ai/blog/generat...
www.deepset.ai/blog/generat...
September 30, 2025 at 3:12 AM
Do you have different models you are trying compare? If you can get more than one metric, I would use a radar. I found this one for RAG from
www.deepset.ai/blog/generat...
www.deepset.ai/blog/generat...
Yeah like little productivity hacks.
I'd like to record the way I work and then and try to replicate it somehow in an LLM. In reality, it's like pulling at threads until you find either how to clean the data or a story. But it involves a lot of dead ends.
I'd like to record the way I work and then and try to replicate it somehow in an LLM. In reality, it's like pulling at threads until you find either how to clean the data or a story. But it involves a lot of dead ends.
September 29, 2025 at 6:55 PM
Yeah like little productivity hacks.
I'd like to record the way I work and then and try to replicate it somehow in an LLM. In reality, it's like pulling at threads until you find either how to clean the data or a story. But it involves a lot of dead ends.
I'd like to record the way I work and then and try to replicate it somehow in an LLM. In reality, it's like pulling at threads until you find either how to clean the data or a story. But it involves a lot of dead ends.
Today I was explaining to someone how I combine 4 columns from 4 different tables to create a single indicator of whether someone is a child.
September 29, 2025 at 5:36 PM
Today I was explaining to someone how I combine 4 columns from 4 different tables to create a single indicator of whether someone is a child.
I often have to try to pull together datasets through detective work... explore columns to see if they are junk or gold. Find the best source/column for measuring something. Most of this is because there is so much data debt in real data systems.
September 29, 2025 at 5:36 PM
I often have to try to pull together datasets through detective work... explore columns to see if they are junk or gold. Find the best source/column for measuring something. Most of this is because there is so much data debt in real data systems.
I like the archeology feeds, e.g. #ancientbluesky and @alisonfisk.bsky.social
September 28, 2025 at 6:34 AM
I like the archeology feeds, e.g. #ancientbluesky and @alisonfisk.bsky.social
And quite a few games are decided by these ridiculous penalties.
September 17, 2025 at 6:25 PM
And quite a few games are decided by these ridiculous penalties.
The chance of scoring from these wide areas of the penalty area from any action is ridiculous low
September 17, 2025 at 6:25 PM
The chance of scoring from these wide areas of the penalty area from any action is ridiculous low