# README

TensorBoard Command

command line tools for TensorBoard application. Download training job results for TensorBoard to visualize them.

Specification

Download Training Job Results

download specified training job's results to a directory for the tensorboard.
- tensorboard' directory path : /mnt/tensorboards/<tensorboard_id>
- training job result's directory path : /mnt/tensorboards/<tensorboard_id>/training_jobs/<training_job_id>
only training job results that belongs to the same TrainingJobDefinition of the tensorboard are allowed to download.
retry if fail to download training job results.
(T.B.D) notify as failed to model_api, if failed after retrying.

Usage

Prerequisite

need to set following environment variables.

export ABEJA_API_URL=https://api.dev.abeja.io
export PLATFORM_AUTH_TOKEN=xxxxx
export ABEJA_ORGANIZATION_ID=1230000000000
export TRAINING_JOB_DEFINITION_NAME=training-job-testing
export TENSORBOARD_ID=1800000000000
# NOTE: these training jobs should be of the same TrainingJobDefinition where tensorboard belongs
export TRAINING_JOB_IDS=1500000000000,1500000000001
# For Debugging
export ABEJA_MOUNT_TARGET_DIR=/tmp/mnt

Command Usage

$ ./abeja-runner tensorboard -h
download user training-results

Usage:
  abeja-runner tensorboard [flags]

Flags:
      --abeja_api_url string                  base url of abeja-api (default "https://api.abeja.io")
      --abeja_organization_id string          identifier of organization
  -h, --help                                  help for tensorboard
      --mount_target_dir string               directory to mount shared file system (default "/mnt")
      --platform_auth_token string            authentication token for platform
      --tensorboard_id string                 identifier of tensorboard
      --training_job_definition_name string   name of training job definition
      --training_job_ids string               comma separated list of identifier of training job

Development

use Docker container environment.

$ docker run --rm -it -v `pwd`:/work -w /work golang:1.12.16-stretch bash

build binary

# prerequisite
$ go install github.com/rakyll/statik@latest

$ make build