Another one of these incredibly interesting online classes came to an end. Machine learning was one of the first two classes that started last fall. As I thought to make AI and ML in parallel would be too much, I decided for AI with the intention to make ML later. The second round of ML was announced for January, but actually started in April. Andrew Ng from Stanford, who some people call a rock star in ML, thought the class. The videos were longer than what I was used to, and I downloaded them to my Android to watch them in the train to work. The homework consisted of review questions and work assignments for octave. The last time I did something with MatLab was more than ten years ago, and I remembered nothing of it.
The class started with gradient descent and logistic regression. And almost everything that followed was compared against and related to them.
I had some prior experience in ML, but no formal training. At TCG I learned the basics from a co worker. Then I implemented a document classification engine using an SVM. I read many books on the topic. Later I develpoed a prediction system for good days and locations for paragliding, again using an SVM as well as an evolutionary optimization. Continue reading “machine learning class”
This weekend was Harley Davidson gathering here in Brunnen. I’m not so much into motorcycles, but I saw some incredibly cool looking bikes. There were a lot of bikes from all over Switzerland, as well as lots from the surrounding countries, and I even saw a bike with a Russian number plate. Saturday they had a big parade, and the whole event was really cool, like a mini open air right around the corner.
I took some pictures:
The raspberry pi, for those living under a rock, is the $25 linux pc that was announced big almost a year ago. It has a 700Mhz ARM CPU, 256MB RAM and an OpenGL ES capable GPU. To enable hardware hacking it comes with lots of GPIO pins. All in all about the performance of a premium smart phone from three years ago. But at $25 !!! The primary focus are school children, and the foundation wants to bring the fun on computing back to the children. Like every geek who read about it, I couldn’t wait to get one. First, the launch was scheduled for September or October, then postponed to February. The foundation decided they would outsource the shipping to some big electronics companies. They told them that a lot of people would try to get one of the first 10’000 boards, but still they weren’t prepared at all. The websites of the pi foundation as well as farnell and rs components were down the entire day. I got up earlier that day, hoping to be amongst the lucky ones. Later I signed up for a pre order somewhere in the queue for an upcoming batch. As with the first batch, each person could still order only one board. Then about a month ago, I received a mail indicating that it was time to place the order. And today I finally received it. Continue reading “Raspberry Pi – at last”
Lots of cameras have a wrong date configured. Mine is even worse: It has a dead internal battery, so it looses the date and time every time I turn it off. When you take a picture, the wrong date is saved to the meta data tags of the resulting photographs. Afterwards, you import them with a photo management tool such as shotwell, and they show up in the wrong place. Sorting doesn’t give you a chronological order. So I was looking for a simple tool to fix the embedded dates. I found exiv2. To install it, type:
sudo apt-get install exiv2
Strangely, it seems to allow only shifting the timestamp and not explicitly setting it. But that’s ok, it is probably the more common case. To adjust the timestamp from 25.7.2009 to 17.6.2012 execute the following:
exiv2 ad -Y3 -O-1 -D-8 ~/Pictures/2009/07/GOPR050*.JPG
Turns out it is also possible to set the date on images that had no EXIF data, although the command is a bit more verbose:
exiv2 -v -M"set Exif.Photo.DateTimeOriginal 2020:09:05 07:17:00" 09/RunAndFly4.jpg
I wanted the robot arm to be a bit autonomous from the computer, and I thought the bifferboard should be powerful enough to drive it. So I wanted to install ROS onto it. My bifferboard runs debian squeeze, and that means it’s not just a matter of installing the packages as with ubuntu. There is a dedicated wiki page about installing ROS on debian, so I was of the opinion that it can’t be that hard. Actually, I once tried to install ros electric already when the bifferboard was still running debian lenny, and I ran into an infinite loop. I was full of hope that this bug had been fixed now with the release of debian squeeze and ros fuerte. But in fact, the first infinite loop was even earlier this time. And after circumventing an issue in pip with the help of a pip dev, I ran into the one I experienced earlier. I was able to find a workaround this time and reported the issue to pyyaml. The bifferboard has no FPU, and from what I can observe in python no inf and strange handling of nan. Continue reading “installing ros on a bifferboard”
Recently, I was searching for a bunch of photographs on all of my harddisks, and couldn’t find them. I know they had been there on the old notebook and on the old computer, and I thought I had copied them over. I could also not find them on any of my backup harddisks. Neither could I find any of the backup DVD’s containing the files that I was searching.
So I wanted to see if I could salvage them from an old hard-disk that I formatted and re-purposed. Actually, I installed a new operating system on that harddrive, but didn’t use it a lot. So, with regular tools that query the filesystem directly, you wouldn’t find a trace. From an eposide of hak5 I remembered that scalpel is a tool for just that. Scalpel is a data forensics tool, that scraps through all the blocks of a raw harddisk, searching for headers and footers of known file formats. This works fine as long as the blocks belonging to a file are arranged linearly, which is not always the case. So I did a quick read up on how it works, and gave it a try.
Here is a good tutorial.
Basically, all I had to do is un-commenting the line with the jpg header definition, and run scalpel on the raw device file (e.g. /dev/sdc1 ) while providing an output folder. That way, thousands of jpeg’s were restored. Lots of them were corrupted due to them not being linearly distributed on the disk. But still lots of files were usable. I’m still looking for the backup DVD’s, but at least I have a fall-back now.
Update 30. July 2014:
If the partition is still intact, testdisk might lead to better results.