It needs a lot of work, but the basics are there. It uses unit selection on a large sample database as well as CMU's arpabet and a few other algorithms.
I'm looking to start work on a deep learning approach to improving the quality in January.
I'm going to spend the next month working on an interactive New Years laser display to project on a skyscraper in downtown Atlanta. I have a bunch of videos of some of the older work I did with lasers in undergrad:
I'm rewriting my engine in Rust now and plan to use four projectors to increase the drawing complexity / capability. It'll be interactive / online, so even those that aren't in Atlanta should be able to interact with it.
I have several courses on Udemy.com. They sometimes go slightly out of date as frameworks release breaking changes.
Right now, I'm just about to upload some updates to my most popular course on React/Redux: https://www.udemy.com/react-redux/?couponCode=HAOW8983
I'm in the middle of your Elixir course right now - it's excellent! I did your first React/Flux course and also the Redux one. Your teaching style works very well for me.
The way I get the most out of it is to pause the video after you outline the project spec, implement the whole thing in my own (bad) way and then watch to see how you do it.
This is quite literally the third time in the past 24hrs that while I've been on the internet looking at other things, your course has come up. Must be a sign... buying it now!
I'm working on a Chrome Extension for YouTube that allows you to seek at any point in a video by a phrase search that matches against the dialogue. It will also allow you to search through the video by image captions. The image captions will be computed for discerning frames of the video using a CNN-RNN model called neuraltalk2.
The front end will be very minimal and so this will allow me to get my hands dirty with React + Redux.
http://www.sensorama.org -- Open Source sensor platform for data science on iOS. I wrote code and did all the design. You get the JSON file of sensor data sampled at 100ms interval (configurable). I get a copy too!. I'm slowly improving it and adding new stuff.
I've been writing this app that lets teachers group kid together based on their performance on an assessment. The idea is that you'd try putting things together in groups that answered questions differently so that they could talk about it and either reinforce their mental models (or test just how robust they are) by teaching a peer or ridding themselves of misconceptions by having a peer help them out. I wrote a GA in Python a few months ago that would take vectors of answers and would group them together and it worked surprisingly well. Now I'm designing the webapp for teachers to orchestrate the whole thing (LMS integration, seeing groups, etc.). I wrote a prototype in Angular 1.x and now that I understand what I'm trying to accomplish really well, it's become my pet project to learn React as I implement a more robust version of it.
I've been working on validating my thought that country music lyrics are all about very similar (and very cliche) topics, and I realized I needed to label all the lyrics I scraped from Genius. Since there are 5000 or so, and I didn't want to have to do all of it myself or rigged with some google doc, I built an app to more easily collect training data. It's generalized to allow for different question types and different documents other than just text as well.
It's also a front in case people want to get in contact for general data scraping or ML needs that I can help with, but the main app is the platform for training data.
I don't have a name for it, which is why it's running with a Heroku url, so suggestions are welcome!
I got curious about how CPU-intensive it would be to run a server for a game like agar.io or diep.io. In these games there are a lot of circles moving around on a 2D map, and you have to check collisions between them.
First was a naive Python version. Test every circle against every other circle for collision, O(N^2). This way I managed only 200 circles on my playing field. I wanted to see how much faster it would be rewritten in C. That got me to 1500 circles.
Now obviously the way I was checking for collisions is silly, some kind of subdivision of space is required to avoid having to check everything against everything. I split the playing field into a grid, and only test collisions between circles in nearby grid cells. That got me to 3000 circles running in Python.
Next I want to write that in C as well, to see how fast it would go.
A Twitter bot called @AmazonArtReview that tries to find funny 1-star reviews of media on Amazon. It's using some crude machine learning and just a fun side project.
Finding humor's tough! I've read some papers on finding humor in reviews, e.g. https://cs224d.stanford.edu/reports/OliveiraLuke.pdf. That's been helpful. It'd be great if I could use the Yelp review set for finding humor as they did but the humor doesn't translate between data sets. Mostly it's still manual curation, which is ok too.
Yeah, that's a good suggestion. I should start storing and using that info. My hypothesis is that humor is inversely proportional to the number f helpful votes. For example, 0 of 57 helpful is probably funnier than 3 of 3.
I'm working on a command line tool that checks Python virtual environments for packages with known security vulnerabilities: https://github.com/pyupio/safety
Rolling my own ergonomic back-up tools atm (implementing incremental, secure backups of MySQL dbs, IMAP Emails and Filesystem) because solutions already exist and I'm a silly person who hates himself and the safety of his data.
I am re-analyzing a library of galaxy merger simulations to estimate how much of the sky is covered by debris pulled off during the tidal interactions between the two galaxies. This may have important implications for interpretation of datasets on the growth lifetimes of supermassive black holes in the centers of galaxies. The simulations were created for a different purpose (matching observations of mergers to numerical models), but are well-suited to this sky-covering project. So I'm essentially mining those simulations in a monte-carlo fashion.
Self hosted, low footprint system health monitoring for the .NET world. Which actually serves as a good test bed for a couple of open-source projects of mine. One released, one isn't yet.
A multithreaded CSV parser which figures out the types of each column. It's part of a larger project, which is basically Excel, but you use SQL since the sheets are put into an actual database. It's useful for when you need to clean up, manipulate, or query spreadsheets.
I know it's probably been done before, but the goal of the project is to familiarize myself with parsers, multithreading, making a C++ project in general, and the newer features of C++11/C++14.
I am working on a project (using Julia) to "learn" an efficient representation of scale invariant many-body quantum states from a microscopic Hamiltonian. The particular structure being learned is supposed to be highly efficient both in representational complexity and computational complexity for predicting physical observables. (Work in progress at https://github.com/sivark/jumera/ )
Also reading/thinking about ideas at the heart of theoretical physics and machine learning. In particular, I've spent the last several months in trying to understand the following two papers in increasing levels of depth
Snapzu Blog Enhancement Suite: It's essentially a voting community widget that you can embed on your blog to make it a lot more social and engaging. This provides any blog with extra content, extra audience, and additional revenue (with revenue sharing). Right now there are similar widgets out there (Twitter, Pinterest, etc) but most are redundant (showing content that's already on your blog) and not really allowing your audience to participate or engage.
Right now were in pre-launch mode (collecting emails of interested bloggers) and are launching all the listed features in about 35 days. This is an add-on to our core community and discussion platform Snapzu (similar to reddit/HN), which has been around for a few years.
A hierarchical 'todo-list'. Aiming to match notepad's ease of use but with a bit more structure and eventual collaborative features. Still very early stage but available here - https://atlas.denchev.io/
Currently adding small tools over it to make an IRC client. The plan is to have iii as the client, wii as the web-protocol-client and then write tii as a terminal app which can either communicate with wii or iii directly.
It's a neat project to learn go and write small simple programs that come together.
The end goal is to also try stream some development and see how that goes. I really enjoy it so far.
I'm working on an agile/kanban like web app. Similar to Trello, Sprintly, Jira etc. I feel like all of those are lacking something - Trello is trying to be too generic, and Sprintly feels a bit "busy" and not as simple as I think it can be done. JIRA is a whole other story of bloat. I'm mostly doing it to actually get to work on a site that's built using Phoenix and Elixir and so far that's been an absolute blast! Decided to throw some SPA in there as well, as that would give me a clear separation of backend and view logic. So decided to take a look at Vue.js, and that has really blown me away as well. So simple to get started with, and felt productive right away! Really having fun doing this :)
About 5 years ago, I had created a bunch of screencasts around OpenGL. I recently (Nov 1st) finally started recording screencasts again.
I'm making super short daily screencasts now called ProgrammingTIL (~1-8 minutes), currently covering React, WebGL, ImmutableJS, Algorithms (CLRS), and Webpack.
So far, I've done ~40 new videos. You can check them out here:
It's a batch file that creates project directories, calls a makefile that builds the thing, and then copies the dlls to the build path. Ugly, and probably wrong in any number of ways I'm unaware of, but it works. Once i'm happy with this, I'm planning to get some basic games cloned in C++ and Unity, using it to generate projects.
Also cloning HN in Hack, but don't expect to see that anytime soon, since I can't find anywhere to host it.
I've been teaching myself web development and recently completed a project I started on that's a website/jobs board for remote jobs: https://remotefriendly.work.
A Lua based Digital Signage tool (written in C/Lua) for the Raspberry Pi. And a complete SaaS platform (mostly written in Python) around that tool, so you don't have to understand neither of them and can get everything up and running in 5 minutes. It's commercially used and growing. And a lot of fun. https://info-beamer.com/pi (the tool/standalone program) and https://info-beamer.com/hosted (the SaaS). Feedback welcome!
I'm reworking my first rails project into Elixir-Phoenix & Vue.js app as a learning exercise. It's a video manager for learning resources (Vimcasts etc.). So it "knows" about parsing a number of feeds, and can download from youtube etc. (shell out to youtube-dl). Allows you to comment and export them all to a page (so you can keep track of what you've looked at and learned).
Quite enjoying reworking it, once it's in a usable state I'll be putting it up on Github and posting a link here. I'm sure there's _lots_ I'm getting wrong, but I'm definitely enjoying the process.
Yes me too. The issue I see ahead is finding work with Phoenix.. I live in a town in New Zealand, and there's no elixir-Phoenix shop that I'm aware of, and I imagine getting remote work without a year or twos production experience won't be easy.
Still, it's fun and I'm looking forward to trying this year's adventofcode.com in elixir :)
I'm also currently working on a web version of Marcus Aurelius "Meditations", mainly because I wanted a way to just randomly show single chapters. https://github.com/ShaneKilkelly/marc
It's been fun so far, especially the process of choosing a freely-available translation, cleaning it up and so forth.
I'm currently writing a book on C# and XML for a publisher as well as working on my new software product Breakneck Install (http://www.breakneckinstall.com). I'm re-writing the UI in WPF so that it gives a better experience since I'm waiting on my code signing certificate to come through so I don't have to worry about dealing with the installer being flagged
Building small app with a friend. It should add some features to Instapaper, Pinboard and Pocket. We are having some trouble with different APIs, but a good thing is that I extracted boilerplate for this kind of things.
Hey thanks! Still really pre-marketing atm as I'm just putting the finishing touches on it. You can ping me an email at hello[at]brandfox.io if you'd like :)
Building http://ossmetrics.com/ - Github Analytics and Lead research platform for open-source businesses.
I started working on it in parallel with my OSS business https://goreplay.org because Github literally does not give you any KPI's. The number of stars, won't tell you anything.
There is so many data generated by open source projects, and if used properly it can give you a huge bonus.
Did not launch yet, but some cool stuff coming soon :)
I've been working on and off on a HTML5 and JavaScript encrypted chat application called Quiet Message. I use the result of a Diffie-Hellman exchange to transmit a key that is used to encrypt messages by the client. It's intended use case is to provide privacy on corporate networks with HTTPS man the middle devices. It's still very basic and messy but I'm slowly making progress.
A management game for arpgs. I don't like spending 8+ hours on a build only to have it die to something stupid in hardcore, so I'm trying to condense the experience into 15 to 20 minutes.
The core gameplay loop is making choices for your build and the ai playing out a few missions/maps at faster than human speed. After the ai's done, you make some more choices with the new loot and experience.
Basically, Football Manager for Path of Exile.
Right now I've got the combat system, DamageMods, StatMods, and status effects.
I'm learning Japanese and wasn't happy with Anki's interface (+ I wanted to learn about Spaced Repetition), so I decided to do Anchor: http://francisco.io/japanese/
Now I'm cleaning it up a bit to allow to easily switch from courses and considering making it into an app. I also want to make a couple of tweaks to the SR algorithm, but it works great already for my needs.
When I was in high school I started a coding club because I really really wanted the environment and culture I found on Hacker News, but in person.
That club had such a profound effect on me, I wanted to make it easier for other high schoolers to create communities of hackers. Hence, I started working on Hack Club :-).
I've been working on a 2D Physics Engine for the folks at Khan Academy to use. It's also a great way for me to learn programming and physics at the same time ^.^
I am making a weight tracking application based off of the hacker's diet[0] using Flask and Pygal. I haven't worked on it in a few days because of school and all, but since I just learned how to process forms, I'll finish it off this weekend.
I'm building https://bandgap.io, which is a platform for computational solid state physics. It computes band structure, Fermi surfaces, and someday, I hope, we can get to computing topological properties. It's done in Django for the backend, Three.js for the visualization, and C++ for the compute.
I've been working on improving my server-side JavaScript - Node/Express etc and basically taking a break from client-side frameworks, React/Angular etc. I miss the simplicity of just building a server-side website.
Anyway, I made a little job board https://wehire.io/ Nothing fancy, but fun to build.
I am working on a Universal Banking app called NuBank. It was previously called giveapenny at http://www.giveapenny.co.
Right now I am in closed beta working on some on cleaning up the interface and also planning on giveapenny.co business app that will help business bank more easily.
Wanted to make some project in Go for a long time. Finally decided to create a 12 factor log router - https://github.com/agnivade/funnel. Progress has been going good and learnt a LOT of stuff along the way. Its been fun !
I've been working on http://www.indeals.co.uk for a couple of years now, a side project that lists food & drink deals for independent venues in 3 cities. We're UK only at the moment but want to keep on growing to cover more places soon!
Got my programming friends doing codewars.com problems, and hacked together a Whatsapp bot that keeps letting the group know everytime someone solves a problem, if someone solves the current problem, and let them know when everyone has solved the current and what is the next one.
It's fun! :)
Climate provides a huge number of command line options for developers to automate their Linux system. Learnt a lot about shell script and various unix tools.
I'm currently working on a real time path tracer for games.
It's pretty neat what a simulation of physically-based lighting enables you to do in terms of lighting and visual phenomena. Needs fast hardware but we're getting there!
Yes, I'm planning to do a series of blog posts about it. In general, Rust is shaping up to be a great language for video games. There are some things that are known problems and that are hard to work with (such as recursive self-referencing data structures) but those are corner cases.
There general tooling of Rust is really nice. Much less so about Vulkan: It has a very limited compiler ecosystem with few optimizers and in general programming GLSL shaders in Vulkan is error prone, hard to debug and you'll be doing a lot of manual loop unrolling.
In my personal time I'm building a nice way to have all my "watch later" videos on youtube in one list, for fast access. At work I am building a react native application that does video streaming.
Both nice and challenging. :)
Going back through history and trying to 'spice it up'. Wars take longer, bigger death counts, longer rail travel rides, Waterloo with dragons, etc. It's fun, if only to read history and then see how to turn it up to 11.
I'm open sourcing one small component from one webapp we have: the pull to refresh feature. It is vanilla javascript, super light and fast, and it won't require you to modify the markup.
http://jungle.horse
It needs a lot of work, but the basics are there. It uses unit selection on a large sample database as well as CMU's arpabet and a few other algorithms.
I'm looking to start work on a deep learning approach to improving the quality in January.
I'm going to spend the next month working on an interactive New Years laser display to project on a skyscraper in downtown Atlanta. I have a bunch of videos of some of the older work I did with lasers in undergrad:
http://lasers.io
I'm rewriting my engine in Rust now and plan to use four projectors to increase the drawing complexity / capability. It'll be interactive / online, so even those that aren't in Atlanta should be able to interact with it.