tensorboard
tensorboard copied to clipboard
Feature: scalar chart export as SVG/LaTeX/pgfplots
Migrated from https://github.com/tensorflow/tensorflow/issues/9024.
@el3ment says:
It would be nice if there was an export option in tensorboard that would export a latex-ready version of the visible graph - perhaps using SVG, Pgfplots or something similar. It's possible now to export CSV, and create the graphic by hand... but having it baked into tensorboard would be nice.
We don't foresee being able to implement this in the near future, but PRs are welcome. Note that the current vz-line-chart component that implements the scalar charts does use SVG elements, but it uses a constellation of such elements that together make up the chart, so even an implementation to export as SVG would require somehow compositing together those elements correctly.
Given that we're already able to export to CSV, couldn't this be roughly implemented by having a "Save to image" option that plots the data from the CSV values? That would at the very least condense the "save to CSV -> load to numpy -> plot -> save to image" sequence of steps into a single step.
seems critical to have plots for publication, otherwise why not just use matplotlib?
would be more performant to use canvas than svg and much easier to save
http://bl.ocks.org/biovisualize/8187844
Are no one working on this feature? As @velocirabbit mentions, having a Save to image option seems rather straight forward given the CSV data, and would be something a lot of us would use frequently.
I have created a simple Python script to help in creating pdf plots using the TensorBoard data for use in LaTeX.
You can find it here: https://github.com/JuliusBairaktaris/TensorBoard-to-PDF
I have created a simple Python script to help in creating pdf plots using the TensorBoard data for use in LaTeX.
You can find it here: https://github.com/JuliusBairaktaris/TensorBoard-to-PDF
I don't mind external tools to do this, but it would be much better if this was an integrated feature into TensorBoard.