In case you have about 10 hours to kill, you should use [Edje Electronics’s] directions to set up TensorFlow on a Raspberry Pi three. In all equity, the period of time you’ll should babysit is about an hour. The remainder of the time is spent constructing issues and also you don’t want to look at it going. You may see a video on the steps required beneath.
You want the Pi with a minimum of a 16 GB SD card and a USB drive with a minimum of 1 GB of free area. This not solely holds the software program, however lets you create a swap file so the Pi can have sufficient digital reminiscence to construct all the pieces required.
The construct step not solely has to create TensorFlow but additionally Bazel, which is Google’s Java-based construct system. There are dependencies between the model of TensorFlow and the model of Bazel so you must ensure the variations match as defined within the video.
The video is predicated on some older directions on GitHub, however these directions aren’t updated and the adjustments required are coated within the video.
Given how lengthy it takes to compile, we puzzled if cross compiling might need been a greater choice. If you happen to want fashions that work with this setup, you possibly can all the time work in your browser.