The hello-world of tensorflow and machine-learning is still kinda hard-core

It took another week, but I finally got around to implimenting the first round of the tensorflow tutorial of MNIST training. And by impliment I mean copy and paste the code sprinkled throughout the tensorflow tutorial webpage. I appreciate that these problems are complicated enough that it probably is best to have hello world come with a fully working code base that explains each section, and it remains fairly concise at 23 lines.

Still, copy and pasting isn’t quite enough

I’m struck by not being satisfied by having copy and pasted the code, where I feel like I’m cheating, but I also don’t really know enough yet to NOT copy and paste code yet. This is ESPECIALLY true in the tensorflow framework. I have seen rather more of scikit-learn code thus far.

I’m looking forward to doing a few more of these, where by then hopefully the framework will have gotten through my thick skull and I can start messing around with the toolset or at least skflow to actually build some of my own models, instead of just repeating known results.

My dumb docker file does indeed work fine and runs tensorflow models

I was reasonably certain this was going to be the case when I could actually import tensorflow last time, but successfully running a model (I have no idea whether it’s fast or slow) did feel really good. I should just get a regular docker and compare cpu to gpu speeds.

---