Brandon Rohrer

@brandonrohrer.com

3723 Followers 2229 Following

Robotics and Reinforcement Learning tinkerer. brandonrohrer.org Wrangler of algorithms. Eater of bread. Sipper of whisky. Reports to a Shih Tzu.

Algorithms and Methods Evolutionary Powell's method for hyperparameter tuning k-nearest neighbors adaptive feature weighting k-nearest neightbors data reduction k-sparse neural network layer Naive Cartographer , a Markov Decision Process learner oknn, an online variant of kNN Pathfinder, a many-to-many shortest path algorithm Sharpened Cosine Similarity, an alternative to convolution for learning 2D features in neural networks Ziptie, an unsupervised feature creation algorithm
A screenshot from the link showing cards for courses on decision trees, polynomial optimization, k-nearest neighbors, multilayer neural networks, autoencoders, and convolutional neural networks
A table containing each proofreader model and its characteristics version model description tokenizer alphabet books error cutoff 01 random_00 random baseline None 0k 0 0.05 02 fomm_00 1st-order Markov 00 20k 10 0.0005 03 somm_00 2nd-order Markov 04 1k 10 0.0005 04 fomm_01 1st-order Markov 00 20k 20 0.0005 05 somm_01 2nd-order Markov 04 1k 20 0.0005 06 somm_02 sparse 2nd-order Markov 00 20k 20 0.0005 07 fomm_02 1st-order Markov 05 20k 416 0.0005 08 somm_03 sparse 2nd-order Markov 05 20k 416 1e-08 09 somm_04 sparse 2nd-order Markov 06 10k 416 1e-08 10 somm_05 sparse 2nd-order Markov 07 5k 416 1e-08 11 fomm_03 sparse 1st-order Markov 05 20k 416 0.0005
model name capitalization punctuation spelling random proof_01 (12) 26 (14) 20 (13) 24 FOMM_00 proof_02 (7) 98 (7) 98 (7) 98 SOMM_00 proof_03 (6) 100 (6) 94 (6) 98 FOMM_01 proof_04 (8) 100 (8) 98 (8) 98 SOMM_01 proof_05 (8) 100 (7) 90 (7) 96
table showing model performance as (precision) recall columns model name capitalization spelling row 1 random proof_01 (9) 16 (12) 22 row 2 FOMM proof_02 (7) 98 (7) 98
Article headline: Now there's an AI tool that adds typos into your emails--so it looks like you didn't use AI
Monthly calendar showing the month of May. Gorgeous full page shot of a dirt trail between wildflowers on a mountain disappearing into trees. 80 point font declaring “ failure is always an option”
A map of where people work in the Boston area shown as overlapping bubbles. Downtown and back Bay are covered heavily and there is a strong representation in Kendall Square, MIT, Harvard, Longwood, and Boston University.
Examples of spelling error detection true positives, false positives, and false negatives.
Tutorials, projects, code, and thoughts collected into topic groups I've generously called Book Projects. Highlights New Releases Being a Staff+ Data Scientist in 2026 Build a custom tokenizer Artisanal Language Models Most Visited Transformers from scratch Setting up an ssh server How to convert RGB color images to grayscale Convolution in one dimension Most Loved Choose your professional path What to do when a leader does something wrong On microsuffering I'm most proud of Solving an easy reinforcement learning problem on hard mode: Inverting a pendulum Naive Cartographer: A Markov Decision Process Learner Ziptie: Learning Useful Features
Being a Staff+ Data Scientist in 2026 Brandon Rohrer | became a data scientist in 2013 when the title was young. It was so new that most companies had no idea what a data scientist should be doing, only that they desperately needed one or they would be left behind. Sound familiar? I've tried to survey the job description of data science a couple of times with varying degrees of success, most recently to go with some informal recommendations for creating data science degree programs. Together with a group of colleages we tried to summarize what data scientists do and the data science subtypes of maker, oracle, detective, generalist. But in the face of changing expectations this doesn't feel like enough anymore. It's time for a refresh. A brief and biased history of the Data Scientist role In the beginning... The field of data science was named in 1997, and the discipline has existed by other names for a very long time. After all, people have been answering questions using data for thousands of years. When data science first got huge, organizations expected data scientists to spin straw into gold—to transform unorganized data archives into profit. Big Data, it was believed, held inherent value, which only needed to be coaxed into cash form. This rarely panned out, so the approach evolved into a) data scientists produce "insights" and then b) "insights" generate profit. This also proved elusive in the end.
A small white dog in a blue coat, standing in front of a towel, covered with lines of treats. Apples, popcorn, carrots, gingerbread cookies, and various others.
12 copies of J.R.R. Tolkien's The Hobbit, each in a different language.
screenshot of the GitHub menu Settings > Copilot > Features with the option "Allow GitHub to use my data for AI model training" set to Disabled
A poster containing the following text (and a QR code linking to the same URL that's given at the end of the text) Learn to see with an engineer’s eyes Professional coaching and mentoring in: Linux | Git | Containerisation | CI/CD | DevOps Cloud Infrastructure | IaC | Testing Engineering Writing “...rigorous, meticulous, and serious in a way that is rare in the unaccredited industry of software engineering, and will give you access to both the technical skills and mindset that will let you outperform other engineers on the market...” - Nikhil Suresh, author of I will fucking piledrive you if you talk about AI again “From now on I'm recommending Iris to everyone who asks me about SWE mentoring. Their future-proofed focus on fundamentals is the best professional advice I've heard. Rock solid.” - Brandon Rohrer, Principal Machine Learning Engineer, Atlassian Book an initial consult: deadsimpletech.com/coaching
A messy nonsense block diagram with hand-drawn colorful squiggles and highlights overlaid.
an antique foot pump pedal mechanism for a sewing machine
Screenshot of iOS icon names being uncomfortably close to each other. Maybe a pixel apart
github.com anthropics / claude-code New issue [Bug] VSCode extension incorrectly attributes manual commits to Claude as author • Open Labels area:ide bug platform: vscode git platform:macos osalmine opened 3 hours ago ... Bug Description Committing manually through VSCode marks the commit as authorder by claude (not co- authored, just completely authorder by it) while it was actually completely authored by me (the user). Claude did not write the code or wasn't even running. Maybe the VSCode extension caused this?
Too many requests
YOU CERTAINLY WILL NOT REGRET CONTINVOUCLY MORGING YOUR CODE
A fast doodle on a blank page, then being reloaded which takes about 2 seconds
A capture of someone (me) drawing "Hi all!" on the Graffiti Wall
A hand-drawn smiley face on a canvas labeled Graffiti Wall
a snapshot of the graffiti wall reading doodle me even on your phone
A very heavily graffitied wall. Sketches and text and scribbles and art.
A pixelized sketch of a very happy raven
A highly pixelated drawing of a butterfly floating over two cute mushrooms
A large empty canvas with the words Graffiti Wall across the top
Tintin captain meme reworked with Wednesday Addams Wednesday: What a week, huh Sergeant? Captain: Wednesday, it's Captain
If you trained a machine learning model (like XGBoost, linear regression, or a neural network) at work last year, what is your title? Machine Learning Engineer 33% Al Engineer 5% Data Scientist 55% Data Analyst 7% 511 votes
If you trained a machine learning model (like XGBoost, linear regression, or a neural network) at work last year, what is your title? Machine Learning Engineer 44% Al Engineer 2% Data Scientist 47% Data Analyst 7% 59 votes • 1w left • Hide results
No one is coming to save you. You are the adult. I'm so sorry. JANUARY 2026
Choose two numbers. Any numbers. -1.21 + 1.2 = -0.01 95 μs painstaking addition 18 ms archival bookkeeping 120 ms round-trip transit to Prague 138 ms for the full experience 689 sums computed in all
Choose two numbers. Any numbers. 3.141592 + 2.7183 = 5.859892 191 μs painstaking addition 28 ms archival bookkeeping 124 ms round-trip transit to Prague 152 ms for the full experience 4 sums computed in all
A small geodesic dome house on a perfect wooden platform in the woods
Man-at-arms from He-man and the masters of the universe
Screenshot from the website: Add two number together. Any numbers. Go ahead. Try it. 5 + 2 = 7
screenshot of post: Moving your code from prototype to production Brandon Rohrer The Blog     RSS So you wrote a cool little weather app that predicts temperatures in your neighborhood. Now you want to be able to share those predictions with everyone of the thousand people who live there. How do you go from "It works on my machine" to "No matter where you are in the world you can use this in your own browser."? There are a lot of dimensions to making your code ready to leave the nest and be on its own in the real world, but an important one is deploying it from your laptop to a globally available server. This post is about one good way to do this. There are lots of others, but this one is solid and, believe it or not, one of the simplest. This deployment starter kit includes Flask, a Python package for turning your code into an app. Gunicorn, a more serious production grade WSGI server. nginx, an industry standard web server. ufw, an Uncomplicated Firewall for Linux systemd the Linux tool for turning your code into a daemon. I won’t bother walking through How to do this because DigitalOcean has an amazing documentation page for setting this up in Ubuntu 22.04. It worked for me on 24.04 as well. Follow their page to the letter and it will get you safely through. And it is such a cool feeling when it does.
ChatGPT 40 › Create an image of the alphabet with each letter represented by a thing, animal, insect or parasite whose name begins with said letter. Image created B Apple H Grape MN Mouse 25 Snake Crab DE Dinosaur Fox L Jnsect Jar King Orange Lion R Quail Pig U Tree Umbrella ) V Yacht Rabbit Parasite parasite Wolf w xX Parasite Z 154 Yaatt Yacht Parasite
A ball point pen embossed with BRACING MYSELF FOR HOLIDAY MERRIMENT
A kind of top hat looking shape that is capable of aperiodic tiling
Bomb-proof your website with mirroring Brandon Rohrer The Blog     RSS In previous posts, we’ve done our best to set up a website that is resilient to traffic, spikes, bots, and mischief of all sorts. ( I, II, III, IV, V VI VII ) But there are still plenty of other things that can go wrong. A good way to protect against this is mirroring. A mirror is like a backup website—not just a backup with the data but of everything, including the server and domain name. I built two additional mirrors for my blog website brandonrohrer.com. The symmetry of triple mode redundancy appeals to me, especially after hearing that it’s a principal used by NASA and growing up on some triples-themed sci-fi. You can take down one, but you still have two backups to work with, plenty of breathing room. If you get really unlucky, two will go down leaving you with a third. Why build one when you can build three at three times the cost? I tried to build in as much redundancy as possible. I got three domain names, each from a different domain name registrar. brandonrohrer.com from Namecheap, based in Phoenix. brandonrohrer.org from Hover, based in Toronto. brandonrohrer.at from Hostpoint, based outside Zurich.
In my last post I said "Automating firewall updates will be an adventure for another day." Today is that day. This is the latest in a growing series on hosting your own static blog. ( I, II, III, IV, V VI VII ) Protecting your website requires you to keep your eyes on it at all times. That can get tiring. Sometimes you need to eat and sleep and maybe work for a living. It can be useful to have automated helpers to fill those gaps. Detection The first step in keeping your blog traffic healthy is to be able to detect when someone is doing something they shouldn't. In my traffic control post I described five behaviors that I wanted to limit Scanning for secrets Fishing for files Rapid-fire requests Trying to hide rapid-fire requests Attempting unsupported actions The post describes these in words, but they need to be made more specific before they can be enforced by an automated detector. Doing this step well is where most of the art and cleverness is required. If you define a rule too narrowly, there will be a lot of IP addresses that should get blocked, but get missed (false negatives for the detector). On the other hand, if you define a rule too broadly, there will be a lot of legitimate IP traffic that gets blocked (false positives) and you will have a lot of sad readers who can't access your cat pictures.
Etsy webpage $69.00 Lobotomy Orbitoclast , Knife & Hammer Surgical Medical Tools This is the Museum quality replica version you will not find anywhere else You will receive 1 orbitoclast pick , 1 lobotomy knife, and 1 hammer
Block an IP address The most straightforward way to block and IP address is in the firewall. It is the tool build specifically for this. To block the address 101.101.101.101, run from the command line sudo ufw insert 1 deny from 101.101.101.101 This instructs ufw (the Uncomplicated FireWall) to insert a rule at the the top of the list (position 1) to deny all incoming traffic from the address. After running this, no restart of the firewall is needed. The rule is active. (ufw docs The position 1 is important because in ufw, the first rule that matches is applied. If there was a rule to allow all addresses that started with 101. and that rule came before the deny rule, then the deny rule would never be reached. While it's possible to block specific ports, or even to block an IP address from seeing particular pages, complex rules and conditions get difficult to analyze very quickly, and can lead to cases where there are loopholes. Use fancy rule combinations sparingly. Parse logs In their raw form access logs are technically human-readable, but they are a lot. I found it really useful to do a little parsing to pull out the bits I'm interested in. (I'm working with the default nginx log format, so adjust this according to your own.) I wrote a script to take advantage of the repeatable structure of these logs to dissect them into their parts. It uses tricks like splitting the log based on brackets and spaces. It productes a pandas dataframe with columns containing the IP address, requested URI, HTTP status code, and every component of the date and time.
Setting: Classic Tintin and Captain Haddock meme, modified Wednesday Addams: What a week, huh Sergeant? Captain: Wednesday, it's Captain
Caring for your webserver Brandon Rohrer The Blog     RSS In parts one through three, we set up a web server, connected it to a domain name, and instituted some basic security. Now comes the fun part! The web server is humming along doing its thing, and we can watch and admire, making little improvements here and there. It’s not required, but for some, this is part of the payoff. A webserver needs weeding and watering It can be counterintuitive that a thing made out of code should need ongoing attention and care. On the surface it seems like it should be self-sufficient, like a wall made of stones. We put the pieces in place where we want them and when we’re happy with it, we stop. It's all silicon and bits after all, why should it need watching? The bigger picture, though, is that the web server operates in a world that’s always changing. Software updates cause tools to behave differently. Edits happen to the HTML and other content we host. There are dramatic changes in who is trying to reach the content and for what purposes. There can be outages, policy changes, and any number of second-order effects in the wider world that can make our web server stop operating the way we want. So on that scale a web server starts to more closely resemble a vegetable garden—something growing, decaying, and very much a product of the environment that it's in. This page is my notes on care and feeding practices I've found helpful and enjoyable. I'm not an expert on this, and this is not authoritative by any means. But I put it here in case you find it helpful. If I missed anything, or got it egregiously wrong, please let me know.
| 403 196.251.87.13 /images/images/cache.php?pass=3a0e007c895888c9c4bf234e2772a0dd | 403 196.251.87.13 /images/images/images/cache.php?pass=331b3elaefcffd452db92a2cicé61c525 lL 403 196.251.87.13 /images/images/images/images/cache.php?pass=26aadf8d261066584120034c6303716F | 403 196.251.87.13 /images/images/images/images/images/cache.php?pass=c93ef613c69e7f82b5446413c46bBaeca lL 403 196.251.87.13 /images/images/images/images/images/images/cache.php?pass=6765bcbc711528e2760d2133cdf17685 | 403 196.251.87.13 /images/images/images/images/images/images/images/cache.php?pass=e2f2d10231c076274cbc76336b530c67 | 403 196.251.87.13 /images/images/images/images/images/images/images/images/cache.php?pass=cfe52c34e32b72e0844fd3ac8c834e96 | 403 196.251.87.13 /images/images/images/images/images/images/images/images/images/cache.php?pass=b4a@cel105e127632d397c11b662c026 lL 403 196.251.87.13 /images/images/images/images/images/images/images/images/images/images/cache.php?pass=1105a5a43bd65cd41776c47401232849 | 403 196.251.87.13 /images/images/images/images/images/images/images/images/images/images/images/cache.php?pass=341a599afb55832de18a18d89cff3cse | 403 196.251.87.13 /wp—-admin/network/network/cache.phn?pass=38efBafddaalead7a?21aef35Fab6d?209
A large hourglass by a pen, notebook, and pair of reading glasses
Git Cheat Sheet Getting Started Start a new repo: git init Clone an existing repo: git clone <url> Prepare to Commit Add untracked file or unstaged changes: git add ‹ file> Add all untracked files and unstaged changes: git add • Choose which parts of a file to stage: git add -p Move file: git mv <old> ‹new>
A plot of bandwidth that shows a precipitous drop (more than nintey percent?) of bandwidth to something like 10 kilobytes per scond usage starting september 9.
18:25 200 185.220.100.250 /multiproc_pipeline.html 18:25 200 185.220.100.250 /multiprocessing.html 118:25 200 185.220.100.250 /numpy_resources.html 118:25 200 185.220.100.250 /one_on_ones.html 18:25 200 185.220.100.250 /open_the_black_box.html 18:26 200 185.220.100.250 /org_response.html 18:26 200 107.189.30.86 /oversimplify.html 18:26 200 107.189.30.86 /pendulum.html 18:26 200 107.189.30.86 /personal_toolbox.html 18:26 200 107.189.30.86 /play_and_record_sounds.html 18:26 200 107.189.30.86 /pocket_guide_data_science.html 18:26 200 107.189.30.86 /professional_path.html 18:26 200 107.189.30.86 /python_data_types.html 18:26 200 107.189.30.86 /python_resources.html 18:26 200 107.189.30.86 /range_normalization.html 18:26 200 45.84.107.174 /rate_of_change.html 18:26 200 45.84.107.174 /regularization.html 18:26 200 45.84.107.174 /rl_noninteger_delay.html 18:26 200 45.84.107.174 /[rss.html 18:26 200 45.84.107.174 /server_naming.html 18:26 200 45.84.107.174 /softmax.html 18:26 200 45.84.107.174 /sql_resources.html 18:26 200 45.84.107.174 /ssh_at_home.html 18:26 200 45.84.107.174 /standard_normalization.html 18:26 200 45.84.107.174 /stats_resources.html 18:26 200 104.244.79.50 /threading.html 18:26 200 104.244.79.50 /transformers.html 18:26 200 104.244.79.50 /uv_cheatsheet.html 18:26 200 104.244.79.50 /vim_md_colors.html 18:26 200 104.244.79.50 /what_nns_learn.html 18:26 200 104.244.79.50 /which_tool_should_i_use.html 18:26 301 104.244.79.50 /ziptie
Part 1 of this series showed how to create a new web server from scratch and Part 2 showed how to set up a new domain name on it. Now it's time to make that website more secure, so it can be protected from the rougher edges of the cyber world. Enable HTTPS Currently the web server we've built is only serving unencrypted (http) requests, rather than encrypted (https) requests. https has become the standard for the web. It's obviously useful when sharing personal information like credit card numbers and love notes, but it's even beneficial for seemingly benign browsing. It's hard to foresee all the ways that someone might use the browsing habits of your readers against them. To protect them, we'll set up https. As with so many topics there is an extremely accessible explanation of how https works in The Wizard Zine on http. If you have been following the other two posts closely, then you are perfectly set up to follow the DigitalOcean tutorial walking through https setup. It builds on all the pieces we've put in place: ufw, a sudo-capable non-root user, a domain name, DNS records, server blocks. I won't bother copying and pasting here. It's an excellent guide so I'll just point you to it.
Attach your web server to a domain name Brandon Rohrer In part one of this series we set up a web server. The next step is to make it act like a real website. Buy a domain name The first thing to do is to name it. Typing http://138.197.69.146 to get to your content is impersonal, boring, and impossible to remember. The name of a website is its domain name. For example, the domain name for English-language Wikipedia is en.wikipedia.org. org is the top level domain (TLD). wikipedia is the second level domain. en is a subdomain. Top level domains come in two flavors, generic and country code. Generic top level domains (gTLDs) include some you are familiar with, like .com and .org, and a bunch you've probably never seen, like .anime and .tattoos. Country code top level domains (ccTLDs) are all two-letter country codes taken from ISO-3166. A few of the most common are .uk (United Kingdom), .de (Germany), and .cn (China). Sometimes people will get creative with country codes and repurpose them. The .ai (Antigua) ccTLD has been co-opted by a whole generation of startups. Same for .io (British Indian Ocean Territory). .me (Montenegro) is a fun one for personalized websites, and the US state of Maine. .it (Italy) and .to (Tonga) are useful for spelling English phrases and words like this.is.it and p.ota.to.
Security measures There are a few things you can do to make your setup more secure against anyone who might want to come and mess around where they're not wanted. Use an ssh key instead of the server account password. Use a passphrase with you ssh key. Use a port other than 22. Use an allow list. Don't allow root login. Enable verbose logging. So you can check on anything hinky. Periodically update sshd.
Planet-size pigeon brains This familiar bird has never gotten much credit for being intelligent. But the reinforcement learning powering the world's most advanced Al systems is far more pigeon than human. By Ben Crair
Hosting your own website Brandon Rohrer Creating a web page is only the beginning. The business of sharing it with the world has a lot of moving parts and a lot of choke points. If an ISP decides to stop traffic to your website or if your hosting platform goes bankrupt there isn't a lot you can do about it. But what you can do is have a solid Plan B — being ready to spin up another copy of your website on another server. A server that you control, or at least a server of your choosing. This guide shows one way to do set up your own hosting. It's not the only way or even the best way. But it is at least a path through the desert. Choose a computer There are lots of cloud providers out there, of all sizes and in a lot of difference countries. Pick one. Or use a computer you have at home, if your ISP will let you get away with serving a website from it. | chose Digital Ocean because | had heard some good things about it and because it wasn't AWS. The first step there is to create an account on Digital Ocean by setting up a "Team", even if it's just you. Add an ssh key Create a public/private key pair on your local machine. Follow the Digital Ocean menus to and follow the instructions there and here are some related docs . It's recommended to create a passphrase for your keys, but don't forget to write it down.
DataBS Conf You successfully placed your order and