We know that Stack Overflow is one of the most useful troubleshooting forums for developers. However, I am grateful and outright elated every time it helps me solve a persistent frustration. Most recently: TensorFlow (for Python) and h2o (for R) refused to load from my scripting IDEs.
At first, I thought it was TensorFlow itself that was not compatible with whatever system I was running. My firewall, maybe. The search term “ModuleNotFoundError TensorFlow” only led me to various GitHub discussions that ended up taking me down an Anaconda version/environment rabbit hole thinking I’d failed outright in installing TensorFlow itself. After trying a few different install modes per forum suggestions, I tabled any thought of Deep Learning with Python. Luckily, the h2o library in R gave more meaningful error messages than the “No module named tensorflow” returned by Spyder.
My hope was that R would have the answers when Python gave me grief, and vice versa. This had been the case several times before, but not this go-around. Error in h2o.init(nthreads = -1) … Two package errors in one session? Gimme a brark. Morning Grind time was running short and it was nearing Morning Commute O’Clock. I tried the first JDK I could find, which was the most up-to-date, and rushed a quick install in time to get one last error before heading off to work. The suspense, it stung.
As a classroom teacher, any downtime I find during the work day is consumed by grading. Plus, with the stress of managing teenagers for 7-8 hours a day, I often need to go for a long walk or hit the gym right after work. It wasn’t until after dinner that I finally calmed down enough to open my computer and have another go. Someone on Stack Overflow had mentioned that h2o was not supported by Java 9, only JDKs 7 or 8. Hope springs eternal! I tried that lead and…no luck. Yet. Further down the thread, someone else had mentioned that they’d needed to fully remove JDK 9 for h2o to kick in. One Stack Overflow post later, it turned out that uninstalling JDK 9 was as simple as deleting version 9 from the JavaVirtualMachines folder in my library.
AAAAAAAAND WE’RE BACK! Not only did this solve my issue with h2o in R, but now TensorFlow loads and I can run my Python models that depend on Keras!
