Overlaying frames-per-second on a benchmark video using R, ffmpeg, and Kdenlive
Feral Interactive is a UK-based porting house that specializes in bringing Windows games to other platforms like Linux and macOS. One of their most recent projects was bringing the game Shadow of the Tomb Raider
to Linux. I wanted to compare the performance of their native Linux version of the game versus running it in Linux using a popular compatibility layer called Wine. Running games on Linux with Wine often incurs some performance cost compared to Windows so there is still a market for native Linux ports that can recover some of that lost performance.
Conveniently, Shadow of the Tomb Raider
contains a built in benchmark tool that will spit out its results to a text file where it can then be analyzed with R. The raw data looks a little like this:
frame time delta memory
1 1 0 0 2341
2 2 14.4 14.4 4462
3 3 35.7 21.3 4462
4 4 53 17.3 4462
5 5 72.1 19.1 4462
6 6 91.6 19.5 4462
Frame is the id of the current frame, time is the milliseconds since the start of the benchmark, and delta is the amount of time it took to draw the frame. Most gamers don't really care about these numbers though; the most relatable metric is frames-per-second which is the number of frames that are able to be drawn in one second. To calculate this I just look a the time it took to draw the previous 50 frames, then 50 divided that time is the rolling FPS.
With FPS calculated, it's easy to use R and ggplot2 to make a nice graph showing the performance of the benchmark over time.
That's neat, but what I really wanted was to overlay the chart over footage of the actual benchmark so that people could see how different in-game scenes effect the frames-per-second. To do this I used a few tools: R again for the chart generation, ffmpeg to turn pictures into a video, and then Kdenlive to edit the video.
To embed a moving chart in a video, I used R and ggplot2 to generate 1 chart per video frame. That works out to 4000 individual charts due to the benchmark being 160 seconds long and wanting 25 frames per second. Each new frame shifts a window showing the next 1/25th of a second of data and 10 seconds worth data over the whole image.
To make things look a bit nicer in the final video, the background of the charts had to be a colour that could easily be chroma keyed out. Chroma keying can remove a certain colour from a video layer, basically green screening. So all 4000 charts looked something like the following beautiful
Turning Charts into a Video:
Thankfully, turning a series of images into a video is rather common problem and there are a lot examples online of using ffmpeg to do this conversion. So I shameless borrowed
the following command to turn all 4000 charts into a video. I won't pretend to know what all of the arguments do, but importantly it is set to 25 frames-per-second to match the timing of the generated charts. Without this the scrolling chart would be too fast or too slow and would not line up with the benchmark footage.
ffmpeg -r 25 -f image2 -start_number 1 -i plots/fps_%d.png -vcodec libx264 -profile:v high444 -crf 0 -pix_fmt yuv420p sottr_fps.mp4 Overlay FPS Video on top of Benchmark:
Kdenlive is an open-source video editor for Linux. Video editing is still one of the areas of desktop Linux that is still a bit lacking, but Kdenlive crucially has a chroma key feature which is the key component in this step. The video generated from the bright green charts is overlayed on the footage of the Tomb Raider benchmark then the chroma key is applied.
In this screenshot you can see the chroma key effect being applied to the bright green of the chart video. It removes the background and turns it into a very nice looking overlay.
So that's it. I really enjoyed this little project because it was the combination of several tools (R, ffmpeg, and Kdenlive) that really made it possible. Each had a specific task and it all came together nicely.
Check out the final result on YouTube
Tags: R benchmarking ffmpeg kdenlive video editting
Bitcoin Prices and Hidden Markov Models
Lately, there’s been a lot of interest in Bitcoin, probably sparked by its almost unbelievable growth in December 2017. However, this past week, we saw the price of Bitcoin drop the just above $6000 which was the lowest it has been since November 2017. So I wanted to take a closer look at Bitcoin prices through the lens of Hidden Markov Models (HMM) to see what conclusions, if any, can be drawn.
Hidden Markov Models are similar to a standard Markov chain model but the where the current state is unknown. Instead of observing the actual state of the process, the only information available is the realization of some other output that is dependent on the current internal state. A somewhat contrived example would be trying to detect whether it is raining, or not, based on how many people you see with umbrellas. The hidden, unobservable state is the weather (raining or not) while the observable, realization of that state is the proportion of people carrying umbrellas (more people carry umbrellas if it’s raining).
Applying this concept to Bitcoin prices, there could be some internal state driving the change in price and different states produce different expected price changes. I assumed that the daily change in price followed a Log-normal distribution, which means that taking the logged value of daily returns should be normally distributed. This made the model slightly easier to interpret. I also used 3 internal states in an attempt to capture bear and bull states with differing volatility.
Below is a chart showing the most likely states during the 2017 and into the 2018 calendar years:
Here each of the three states are coloured. The blue state was characterized by positive average returns and low volatility. The red state also had generally positive returns but higher volatility. Finally, the green state had mostly negative returns and also high volatility.
I also ran a quick Shapiro-Wilk test on the log-valued daily returns which was unable to reject the null that daily returns come from a normal distribution. This means that there wasn’t enough evidence to disprove the assumption that price changes follow a Log-normal distribution.
This is all fine and good, but what would be really cool is if the fitted model could be used to predict the future price of Bitcoin. So I ran 10,000 30-day simulations to get an expected future price and a confidence interval. This is what it looks like:
This shows the predicted Bitcoin price, and the actual price change during the prediction interval. The shaded regions also represent the 95% and 80% confidence intervals, based on the 10,000 simulations. In this instance, the HMM was not exactly a great predictor. Bitcoin has been incredibly volatile and I think it’s extremely difficult to make any meaningful predictions using closing price alone.
If you’re interested in taking a closer look at the R code used to fit the HMM model and generate the charts, you can find it on my Github
Tags: R bitcoin finance hidden markov models