Awesome

UniversalTensorBoard

Deprecated in favor of https://github.com/PhilipVinc/TensorBoardLogger.jl/

This is a proof of concept for using TensorBoard for arbitrary numeric logging. You could use it with Flux.jl, or with what ever else you want.

Won't work on Windows, because TensorFlow.jl does't work on Windows

Usage

usage is very simple

• @tb_log foo to log the current value of foo.
• set_tb_logdir(new_dir) sets the logging directory to new_dir
• reset_tb_logs() clears all current logs
• tensorboard() opens TensorBoard in your default browser (or you can open it manually the normal way)

Example

Scalars

using UniversalTensorBoard

best_ele = -1
total = 0
for (ii, ele) in enumerate(50rand(1_000))
    if ele > best_ele
        best_ele=ele
        @tb_log best_ele
    end
    total+=ele
    running_mean = total/ii
    @tb_log running_mean
end

Vectors -> Histograms

using UniversalTensorBoard

for ii in 0:0.2:3
    @tb_log randn(100)+ii
end