TensorBoard Visualisation about self-driving-car

[RyanZotti]

@zfhall Great question. Short answer is that my TensorBoard code is broken because I had started to dismantle it. I used to have working TensorBoard functionality, but back when I first started the project Google kept changing the TensorBoard API and it was a pain to keep the TensorBoard code updated. And on the other hand I felt there wasn't much to gain from TensorBoard. I never really found it that helpful, so now I just write my accuracy to a simple: text file:

with open(self.results_file, 'a') as f:
    f.write(message.format(epoch, train_accuracy, test_accuracy)+'\n')

What do you want to use TensorBoard for in particular? It's possible Google added useful functionality to it since I last looked at it, but my working TensorBoard code is buried waaayyy back in my commit history, so it'll take me a while to find and test it.

[zfhall]

@RyanZotti That makes sense. Many of the tutorials I have looked at seem to use outdated code for TensorBoard and I have ran into many problems because of this.

I wanted to use TensorBoard to visualise accuracies , and cross entropy losses. Mostly because I prefer things visualised and work with plots on a daily basis so its just the way I like to do things. I also don't plan on putting this project to bed any time soon so its a nice way to expand on what has already been done.

Please don't go out of your way to go through your history though. I'm just very new to TensorBoard so with some more research I'm sure ill be able to figure it out.

[RyanZotti]

@zfhall If all you're trying to do is visualize the accuracy / loss, then writing to a file and using something like python's Seaborn Seaborn or Matplotlib would probably be easier and less prone to changing code. Both libraries have been around much longer than TensorFlow and have much more stable code examples on Google.

[zfhall]

@RyanZotti Thanks for the info. I think ill just use matplotlib as i'm more familiar with it and your right it will be easier than this tensor board stuff!