Skip to content
Snippets Groups Projects
Select Git revision
  • aa8d3ff98699476757b88ef52085e2d7ac76ad08
  • master default protected
  • emoUS
  • add_default_vectorizer_and_pretrained_loading
  • clean_code
  • readme
  • issue127
  • generalized_action_dicts
  • ppo_num_dialogues
  • crossowoz_ddpt
  • issue_114
  • robust_masking_feature
  • scgpt_exp
  • e2e-soloist
  • convlab_exp
  • change_system_act_in_env
  • pre-training
  • nlg-scgpt
  • remapping_actions
  • soloist
20 results

README.md

Blame
  • Code owners
    Assign users and groups as approvers for specific file changes. Learn more.
    README.md 1.17 KiB

    Plot training curve

    Plot training curve directly from tensorboard event files by using for instance

    python plot_results/plot.py --dir=EXP_DIR --map-file=example_map.json --max-dialogues=MAX_DIALOGUES --out-file=plot_results/

    The structure of the map-file is like this:

    [
      {
        "dir": "dir_1",
        "legend": "Algorithm1"
      },
      {
        "dir": "dir_2",
        "legend": "Algorithm2"
      }
    ]

    The value of legend will be the depicted in the plot legend.

    The file structure of the exp_dir is like this:

    ├── exp_dir                  
        └── map["dir_1"]
            └── experiment_seed0*
                └── tb_dir
                    └── events.*
            └── experiment_seed1*
                └── tb_dir
                    └── events.* 
        └── map["dir_2"]
            └── experiment_seed0*
                └── tb_dir
                    └── events.*
            └── experiment_seed1*
                └── tb_dir
                    └── events.* 

    If you want to truncate the figure to a certain number of training dialogues on the x-axis, use the argument --max-dialogues.