Tensorboard调试及页签隐藏

调试：

        tensorboard入口文件 ./tensorboard/main.py    function:run_main()

def run_main():
  """Initializes flags and calls main()."""
  program.setup_environment()
  tensorboard = program.TensorBoard(default.get_plugins(),
                                    default.get_assets_zip_provider())
  # 增加启动参数
  sys.argv.append('--logdir='+eventpath)
  # 重新配置tensorboard启动项
  tensorboard.configure(sys.argv)
  try:
    from absl import app
    # Import this to check that app.run() will accept the flags_parser argument.
    from absl.flags import argparse_flags
    app.run(tensorboard.main, flags_parser=tensorboard.configure)
    raise AssertionError("absl.app.run() shouldn't return")
  except ImportError:
    pass
  tensorboard.configure(sys.argv)
  sys.exit(tensorboard.main())

运行main.py，程序和shell启动相同

 

页签管理：

        项目中经常会有需求对tensorboard的eventlog页签进行相应的管理，比如image中有部分图片不想让用户直接查看

tensorboard的文件解析由各自的plugin处理：

<tensorboard.plugins.core.core_plugin.CorePlugin>

<tensorboard.plugins.beholder.beholder_plugin.BeholderPlugin>

<tensorboard.plugins.scalar.scalars_plugin.ScalarsPlugin>

<tensorboard.plugins.custom_scalar.custom_scalars_plugin.CustomScalarsPlugin>

<tensorboard.plugins.image.images_plugin.ImagesPlugin>

<tensorboard.plugins.audio.audio_plugin.AudioPlugin>

<tensorboard.plugins.graph.graphs_plugin.GraphsPlugin>

<tensorboard.plugins.distribution.distributions_plugin.DistributionsPlugin>

<tensorboard.plugins.histogram.histograms_plugin.HistogramsPlugin>

<tensorboard.plugins.pr_curve.pr_curves_plugin.PrCurvesPlugin>

<tensorboard.plugins.projector.projector_plugin.ProjectorPlugin>

<tensorboard.plugins.text.text_plugin.TextPlugin>

<tensorboard.plugins.interactive_inference.interactive_inference_plugin.InteractiveInferencePlugin>

<tensorboard.plugins.profile.profile_plugin.ProfilePlugin>

./tensorboard/backend/application.py 

function: standard_tensorboard_wsgi(flags, plugin_loaders, assets_zip_provider)

  plugins = []
  for loader in plugin_loaders:
    plugin = loader.load(context)
    if plugin is None:
      continue
    # 可以根据自己的策略进行过滤 此处过滤ImagesPlugin
    if plugin.plugin_name == 'images':
      continue

    plugins.append(plugin)
    plugin_name_to_instance[plugin.plugin_name] = plugin
  return TensorBoardWSGIApp(flags.logdir, plugins, loading_multiplexer,
                            reload_interval, flags.path_prefix)

原网页：

过滤ImagesPlugin后