Imported from the Raspberry Pi Foundation website

On May 9th I wrote about a broken Tensorflow package on RPI0W using Python 3.7.3. Here I will give more specifics to the problem. I was using Raspbian Stretch, latest update with manually downloaded, compiled, installed Python 3.7.3. The version of Tensorflow I attempted to install and execute was 1.13.1.

After the failure, I posted a github broken package report. The error was pushed upstream and confirmed by the tensorflow people. This proceeded smoothly. A new ARM package was released for Python 3.7 as a result.

Earlier this week I attempted to install and execute the new package, but now have come up with a new set of errors. One is that tensorflow-estimator (1.14.0rc1) is loaded as a dependency instead of (1.13.0) as was loaded previously. I have never known a release candidate used as a dependency in place of a known stable production version. The second error (maybe related to the first) is that Tensorflow no longer loads a valid H5 model file–one that I have been using for months with the same version of Tensorflow in Python 3.6.8.

I filed a new bug report on github for this issue. Lets see where this goes. In the meantime I guess I will continue to use Python 3.6.8.