This package provides a julia wrapper for VisualDL, which is a deep learning visualization tool that can help design deep learning jobs.
Currently, the wrapper is written on top of the Python SDK of VisualDL by PyCall. I have tried to write the wrapper on top of the C++ SDK by leveraging CxxWrap.jl. But unluckily a strange error encountered. Hopefully I'll figured it out later and swap the backend into C++.
First, install the Python client of VisualDL. Checkout here for a detailed guide.
Then add this package as a dependent(only tested on Julia v0.7).
(v0.7) pkg> add VisualDL
First, initial the logger.
using VisualDL train_logger = VisualDLLogger("tmp", 1, "train") test_logger = as_mode(train_logger, "test")
for i in 1:100 with_logger(train_logger) do @log_scalar s0=(i,rand()) s1=(i, rand()) end with_logger(test_logger) do @log_scalar s0=(i,rand()) s1=(i, rand()) end end
for i in 1:100 with_logger(train_logger) do @log_histogram h0=(i, randn(100)) end end
for i in 1:100 with_logger(train_logger) do @log_text t0=(i, "This is test " * string(i)) end end
for i in 1:100 with_logger(train_logger) do @log_image i0=([3,3,3], rand(27) * 255) end end for i in 1:100 with_logger(test_logger) do @log_image image0=rand(10, 10, 3) * 255 end end # force save and sync save(train_logger) save(test_logger)
visualDL --logdir ./tmp in current dir. Then launch the visualdl service and watch the above pictures in browser. The default url is
LogReader~~ and tests
about 1 year ago