Here’s the short version: I’ve made a Creative Commons licensed version of the Galaxy Brain meme images. I’m releasing these images under the CC-0 license, which means you can use them in just about anything, including commercial works. I’ve made versions with both a masculine and feminine figure. More notes below!
We made these because we’re in the middle of writing Head First Swift, to be published by O’Reilly Media. The Head First series of books is intended to be a lighthearted and fun way to teach technical content, and my co-author, Paris, asked if we could add a galaxy brain meme in the book. Unfortunately, the nature of memes means that not only do we not have permission to publish the material in a book, it’s next to impossible to try and track down the rights-holders for those images. So, I made my own.
While I was making them, I made a couple of changes to the commonly-used formula. The initial image in most galaxy brain sequences shows an x-ray image with a picture of a scaled-down brain; I’ve always found this to be pretty ableist, especially since the meme is frequently deployed to make fun of people with bad opinions. Bad opinions are rarely the result of brain physiology. So, I opted to make the first image simply a non-glowing brain.
After posting the images on Twitter, I received several requests for a version of the images that didn’t just use the standard masculine model. Hot takes aren’t the exclusive domain of men, so I put together an additional set. (When I made these, I was glad that I didn’t make the first image use a small brain, because that would have made that image much more likely to be used in isolation as a sexist trope.)
Thanks for reading this far, and I hope these pictures bring you joy!
It’s a bit tricky to see where to begin with OpenAI’s (in)famous GPT-2 model. This blog post is our first in a small series about NLP. We hope it helps!
Our preferred way of installing and managing Python, particularly for machine learning tasks, is to use the Anaconda Environment.
⚠️ Anaconda’s environments don’t quite work like virtualenv, or other Python environment systems that you might be familiar with. They don’t store things in the location you specify, they store things in the system-wide Anaconda directory (e.g. on macOS in “/Users/USER/anaconda3/envs/”). Just remember that once you activate them, all commands are inside the environment.
Anaconda bundles a package manager, an environment manager, and a variety of other tools that make using and managing Python environments easier.
➡️ While inside the activated Conda environment you’re using, install the requirements specified inside the GPT-2 repository:
pip install --upgrade -r requirements.txt
➡️ Use the model downloading script that comes with the GPT-2 repository to download a pre-trained model:
python3 download_model.py 774M
⚠️ You can specify which of the various GPT-2 models you want to download. Above, we’re requesting a download of the 774 million parameter model, which is about 3.1 gigabytes. You can also request the smaller 124M or 355M models, or the much larger 1.5B model.
You might need to wait a little while as the script downloads the model. You’ll see something like this:
This will fire up the model, allowing you to enter some text. You’ll eventually (likely after some warnings that you can ignore about your CPU and the version of TensorFlow) see something like this:
Model prompt >>>
Enter some text and press return (we recommend only a sentence or two), then wait a bit and see what the model generates in response. This could take a while, depending on your computer’s capabilities.
We’ll be back with a follow-up article, exploring how you can actually use GPT-2 for something useful. Stay tuned!
This is the first post in a two-part series. Stay tuned for the second part.
One of the projects that we’re most excited about is our ongoing work to bring Night in the Woods to iOS. This has been an enormous amount of technical effort, and has led to a bunch of very cool spin-off projects – for example, Yarn Spinner only exists because we built it for Night in the Woods, and the research we’ve been doing in sprite compression to fit the game into a tiny amount of memory has been a blast.
One of the things we’re doing right now is improving the performance of the game on the device. Just like when you’re building a game for PCs or consoles, you want your game to be running at a high, stable frame rate. For mobile devices, this is a little more complex – you also have to consider the impact that the game has on the phone’s battery, because nobody wants to play a game and then find that their phone’s about to die.
Another important element of performance for mobile games is thermal impact. When the phone’s running a game, it’s making the hardware do a lot of work, and this makes it heat up. Modern systems – that is, anything made in the last 40 years – detect when they’re under thermal stress, and reduce their performance to avoid damage. This means that you might be able to achieve a solid 60 frames per second when you start playing, but that might get laggy in a couple minutes (or seconds!) of play. (Thermal stress is also a consideration for PCs and consoles, but it’s more critical for phones due to the reduced amount of space inside the chassis, the lack of active cooling, and the fact that you’re holding it in your hands.)
Spotting the Problem
The first thing that we noticed was that the reported frames per second in the game was low.
(We use an FPS counter called Graphy to visualise FPS. Graphy’s great – it’s easy to set up, reliable, low-impact, and open source. You should use it!)
The FPS counter was reporting an FPS count of about 20 to 40 frames per second on my iPhone X, depending on the scene. The other thing that was spotted was that the frame rate was inconsistent to boot.
What was going on? Night in the Woods, despite its gorgeous visual style, doesn’t actually have hugely complex scenes. When diagnosing a problem like this, the first thing to do is to figure out where the bottleneck is – on the CPU, or the GPU.
Finding the Bottleneck
Xcode has a very simple tool for checking which part of the system is under the heaviest load. When running the game from Xcode, you can just click on the Debug navigator, which will show you a summary of the app’s performance.
The CPU usage and energy impact here are pretty high. Not enormously high, considering it’s a game, but still high. As the game itself reports, it’s barely achieving 30FPS frame rate. When we select the FPS element, we get some more detail:
The FPS report is showing that the device is 100% utilised, and that the CPU and GPU are both taking a full 28 milliseconds to produce the frames. The renderer is under significantly more load than the tiler, which effectively means that most of the work is being done by the fragment shaders, and not by having to deal with a lot of geometry (something that we’d expect, given that the polygon count of Night in the Woods, like most other 2D games, is not terribly high.)
Even though the CPU and GPU times were the same, this doesn’t necessarily mean that they’re under the same degree of load. Switching over to the CPU report showed that the CPU wasn’t really at maximum load (though this can be deceiving, since it may have been the case that a single core was at max load.)
The analysis that we’d seen so far seemed to suggest that the GPU might be the best place to look at, so we pulled out one of the two biggest and baddest tools in the chest: Instruments.
Instruments can show a huge amount of data about the performance of an app. Happily, recent versions of the app come with the Game Performance template, which sets up a number of recorders that relate to how a game performs.
So, we ran our scene while recording data, and got back a simply enormous amount of data. It’s actually rather pretty.
There’s a lot of charts here, but the key thing to look at is at the bottom of the window, where the GPU state and display information is shown. Let’s zoom in on that.
The purple bar at the top shows when the GPU was busy doing something. The bar is entirely filled, which means that there was no point in the run when the GPU was allowed to be idle. That’s bad, because when the GPU is active, it’s pulling energy out of the battery, and heating up the phone. We already knew this, because Xcode’s report was showing us that the device utilisation was 100%, but it’s nice to see this in a little more detail.
Where we start to see more useful information is in the ‘Display’ instrument. The Display instrument shows four charts:
The current frame shown on the screen
When a new frame was delivered to the screen, ready to be shown
When the screen vsync’ed (that is, when it attempted to swap to the next available frame)
Any frame stutters that Instruments found. There aren’t any in this screenshot, but that doesn’t mean that the display isn’t stuttering.
Take a close look at the top row of the Displays instrument. It’s showing the duration of each frame on screen, and they’re not all the same. Some are 33 milliseconds, some are 16. That’s what’s causing the janky appearance.
What’s interesting about this is that the frames are taking the same amount of time to be produced! We know this because the second row, labelled ‘scaler’, shows that each frame is being submitted to the GPU after about the same amount of time. The reason why some frames list for 33 milliseconds and some last for 16 is due to the fixed rate at which the display is swapping to the next available frame.
Understanding Frame Judder
To figure this out, let’s look at the timing information for four frames.
Here’s what’s happening at each of these four steps.
The blue frame is on screen. The next frame, shown in green, is submitted to the display.
The display hits its next vsync point. The green frame is ready, so it’s shown to the user.
The orange frame is submitted to the screen, but it’s just too late for vsync. The screen therefore has no new frame to show, so it keeps showing the green frame for another sync interval. The orange frame is eventually shown at the next vsync.
Meanwhile, the blue frame was being drawn, and it’s ready to go right away. It’s submitted to the display, and drawn right after that. The orange frame was only on screen for a single sync interval.
The problem here is not that the frames aren’t taking too long to draw. The problem is that the game is trying to send them to the screen at too high a rate. Ironically, in order to make the game smoother, we need to reduce the frame rate.
This was a one line change:
Application.targetFrameRate = 30;
The result was this:
There are two things to note here.
First, the purple area now has gaps in it, indicating periods of time when the GPU was not active. This is a good thing! Less energy being drawn from the battery, and less heat.
Secondly, all of the frames are on screen for a consistent amount of time. They’re never missing a vsync, because the game isn’t trying to get them onto the screen as fast as it can. The game can now focus on operating at a consistent 30 frames per second.
The result is lower power consumption, and a smoother gameplay experience.
We’re Not Done Yet
But this was only part of the problem. As I mentioned earlier, the scenes in Night in the Woods are not incredibly complex; why was it taking so long to produce the frames? To solve that question, we needed to take a much closer look at the internals of how the GPU was drawing each frame.
We’ll look at how we achieved even better results in the next post. In the meantime, I’ll tease you with this:
The popular open source narrative game development framework, Yarn Spinner, which is maintained by Secret Lab and a fabulous community, has reached version 1.0. As part of our 1.0 release, we’ve debuted 5 exciting new features:
Compiled Scripts — Yarn Spinner now compiles to a binary format.
Automatic Compiling — In Unity, your Yarn scripts will automatically be compiled when they change.
Line Tagging — You can automatically add unique tags to lines of dialogue, and generate a .csv file to send to translators with the click of a button.
At Velocity Berlin 2019, we were asked to give a talk to a crowd of largely non-game developers, on how games make gather and analyse telemetry in order to improve the experience of players. In this post, we’ll recap what we talked about, and provide links for further reading!
This talk was largely a literature review of advice and techniques from game developers, and we’re tremendously grateful for their generous sharing of knowledge. In particular, three stand-out talks from GDC were hugely useful:
We’d also like to thank Tony Albrecht from Riot Games, whose advice helped us put this talk together.
In order to talk about how games use data, we’ll break the discussion down into four main topics: what data is gathered, how that data is gathered, how the data is analysed, and how the changes are deployed.
What Data is Gathered
Much of the data that games gather is not unique to games. Just about every product out there collects data on when it’s launched, how long the session lasts for, and how much interaction the user has with it; by gathering this data, it becomes possible to understand patterns of usage in terms of sessions.
There are three critical metrics that need to be gathered in order to gain a rough understanding of how a game is used:
Session duration: how much time elapses between the user starting and finishing a stretch of interaction.
Session interval: how long the user waits before starting another session.
Session depth: how many interactions the user has in each session.
It’s important to note that you can’t analyse these individually. If the average session interval is decreasing, that may indicate that people are coming back to play more and more, but it could also mean that players are opening the game, checking to see if there’s anything new, and then immediately leaving.
By gathering this data, it becomes possible to get a picture of the number of daily, weekly and monthly active users for the game. These figures represent how many unique users had a session in the game over the specified period; generally, you want this to be going up over time, though players tend to drop off a game over time. Daily active users counts tend to be quite spiky, because of players who hear about new content and updates and jump into the game, but don’t stick around for long. As a result, monthly active users tends to be the main reported figure, because it effectively smoothes out trends in player population; in the 2018 annual report for Activision Blizzard, one of the key figures that they highlighted was MAU across Activision, Blizzard and King.
In addition to these standard metrics, there are also data points that are unique to games. These tend to vary based on the type of game, but generally include things like score, death, level number, and position. Position is a particularly interesting one, because it’s very easy to visualise and plot against other key events – we’ll come back to this in a moment.
However, these concrete measurements of player behaviour aren’t good at getting an understanding of whether the player enjoyed themselves in the game. To fix this gap, Call of Duty: World War II directly asks their players if they had fun, using a single yes/no question that’s designed to minimise the burden of answering. Interestingly, the development team reports an average non-skip rate of between 60-80%, even when the answer order randomisation places the option to skip as the default. This is significantly higher than they expect.
Finally, games typically record performance data on how smoothly the game is running. Games typically aim to play at 60 frames per second, which means that each frame has only 16.6 milliseconds to render. When you have only about a dozen milliseconds to render, every one of them counts.
As a result, League of Legends records two kinds of performance data: first, regular telemetry reports are sent during a game, giving an idea of the impact that the most recent patch has had on performance. Additionally, the game records performance data for each frame as it’s played, and then compresses and uploads the data at the end of a round.
Data collection is the process of delivering the data to the developer for analysis. There are a few ways to do it, and a great talk from Tom Mathews from 343 on how they built their telemetry systems for Halo 5 is a great place to look at. Some highlights of his talk include the fact that they converted their logging system away from unstructured logging strings to a formal, schema-based format based on Microsoft’s Bond format.
Among a few other benefits, this logging system allowed for sub-second response to telemetry, which meant that the game itself is able to respond to the data. As a result, in-game elements like end-of-game reports and leaderboards are driven by the telemetry system, rather than having to build a separate system for this purpose.
How the Data is Analysed
The data received from games can be divided into two main categories: spatial, and non-spatial. Spatial game data is anything that’s related to the player’s position in the game, whereas non-spatial data is everything else, and includes data like in-game performance, skill, and time spent in the game.
Non-spatial data is generally used to get a picture of how players are enjoying the game, and to generate a predictive understanding of whether players are coming back to play more. The fun survey from Call of Duty: World War II is particularly interesting, because they gather data on whether the player enjoyed the game or not – the developers don’t have to infer this data from simpler things like the fact that players are coming back.
This allows them to link player fun to other variables, with some surprising results. As might be expected, in-game performance – that is, how many kills a player got versus how many deaths they had – is a strong predictor of how much fun the player had in the game, but it’s not the only important variable. Player tenure (the total duration of time spent in game, across all sessions) and player skill (total kills versus total deaths, across all sessions) were found to be strong predictors of players reporting that they didn’t have fun. The Call of Duty development team’s theory for this is that the longer a person plays a game, the more critical of it they become; additionally, they found that player fun reports drop off once a player’s skill level becomes greater than that of the median player.
A particularly interesting note made by the development team was that the margin by which a team won was less impactful on fun score than the margin by which the team lost by. That is, a player whose team won by 50 points was just as likely to say that they had fun as a player who won by 5 points. This wasn’t the case for the losing team, however – a player whose team lost by a large margin was much more likely to report that they didn’t have fun than a player who only lost by a few points. This is good empirical evidence for the widely held belief that games that end in close ties are better for everyone.
In the area of spatial data, an excellent paper by Anders Drachen and Matthias Schubert, “Spatial game analytics and visualization” (PDF) proposes four primary types of spatial data analysis: univariate/bivariate, multivariate, trajectory, and behavioural analysis.
Univariate/bivariate analysis focuses on either one or two variables, in which one of those variables is player position. This is usually seen in the form of heat maps for levels, which allow developers to get a good understanding of where players die in levels. For example, consider this heat map for the level de_dust2, from Counter-Strike. Red areas indicate places where lots of players die.
Some interesting observations from this heat map:
Corridors and doorways are hotspots for player death
Doorways exhibit diagonal lines of player death, indicating where players lie in wait
Certain long lines of player death indicate places where players have long sight lines
A grid pattern can be seen at the bottom and top areas; these are the player spawn locations, and represent players starting the game there, deciding they don’t like their team, and quitting.
An excellent discussion on using heat maps to analyse level flow and gameplay balance is Sean Houghton’s post Balance and Flow Maps, which discusses an analysis of gameplay balance in maps used in Transformers: War for Cybertron.
Multivariate analysis allows for some more complex and sophisticated analysis of level content. Georg Zoeller’s excellent 2011 talk at GDC about the analysis tools used in Star Wars: The Old Republic shows how they combine information about player death with the locations and levels of monsters in the level, which were used to figure out balance problems with the level progression curve.
Trajectory information can be used to plot the movement of individual players through the game’s space, and is useful for detecting outliers and unintended paths through the environment. Jonathan Dankoff’s post on using trajectory analysis in Assassin’s Creed Brotherhood highlights how players could bypass tutorial content by not following the expected path through the environment.
Finally, spatial information can be used to derive data about player behaviour in the game. Mahlmann et al’s paper, “Predicting player behavior in Tomb Raider: Underworld” (PDF) was able to use information like how long players spent in the early parts of the game to predict how far through the game they’d get before they gave up – something that’s extremely useful in balancing game difficulty and production investment in the game’s content.
How Changes are Deployed
When a game has made changes, it’s time to get the updated version out to players. There are a variety of ways to do this, with varying levels of disruption to players.
The most straightforward way of doing it is to release a new version of the game via digital distribution – that is, via Steam, itch, and the various App Stores. This is conceptually simple, but has a few downsides: players may not be aware of the update, or may choose not to update, which fragments the installed player base. Additionally, the size of the updates may be large, which reduces the chance of all players updating.
The local patching model adopted by the Nintendo Switch is quite interesting: players who want to form a local network and play a game are able to compare their installed version, figure out who has the most recent update, and then distribute the patch locally, without relying on internet access. This feature is especially important when you consider that one of the main marketing points of the Switch is the ability pick it up and take it outside; the Switch has no cellular internet connectivity, so local wireless communication is all it has.
If a game is designed to be competitive, opt-in beta streams can be used. In this model, players choose to receive beta versions of patches, and play the game in a testing mode. Because game changes frequently change the balance of play, players who want to maintain a competitive edge have an incentive to play the beta stream, and accept the risk of bugs and data loss.
Game changes don’t necessarily require updates to the code or assets, and small hot fixes can be applied. Some notable games that do this include Borderlands and Fortnite, which download a small patch on every game load that tunes gameplay content. These patches are typically only kept in memory, and are lost when the game exits; hot fixes are generally rolled up into a permanent patch after some time.
In order to minimise downtime, a blue-green deployment model for server updates is frequently common. When a new patch becomes available, existing servers are kept online for as long as there are players connected. All new players connect to servers running the latest version, and older servers are shut down as players disconnect from them. This means that players aren’t required to leave the game when a new patch lands; however, this model only works in games where players are separated into discrete sessions, and doesn’t work in single, shared-world environments like massively multiplayer games. For example, Star Wars: The Old Republic shuts down every Tuesday night for a few hours for patch deployment, and all players are kicked from the server.
We had a great time presenting this to a room full of operations and deployment experts, and we feel that there’s a lot that games can bring to the wider world of operations management. The video recording of our session will be available soon, and we’ll add it to this post when it arrives.