List runs

List runs

You can list runs (sorted by performance) using the command python main.py ls. It has the following options

  -m MODEL, --model MODEL
                        Only show certain models (default: None)
  -r RUN_PATTERN, --run-pattern RUN_PATTERN
                        Filter by run name pattern (default: None)
  -f FILENAME_PATTERN, --filename-pattern FILENAME_PATTERN
                        Filter by name of training data input file (default: None)
  -p PARAMS [PARAMS ...], --params PARAMS [PARAMS ...]
                        Display certain hyperparameters instead of default ones (default: None)
  -a {micro,macro,weighted}, --averaging {micro,macro,weighted}
                        Precision/recall/f1 averaging mode (default: macro)
  -t TOP, --top TOP     Maximum number of models to show, use -1 to show all (default: 40)
  --metrics {accuracy,f1,precision,recall} [{accuracy,f1,precision,recall} ...]
                        Metrics to display (also defines sorting order) (default: ['f1', 'precision', 'recall', 'accuracy'])
  --names-only          Only list names (default: False)
  --all-params          Show all params (default: False)

By default the runs are sorted by f1_macro performance.

Example

In order to show the top 60 runs from a grid search with the partial training run grid_v1. Additionally we would like to show the grid searched parameters (min_num_tokens, min_num_chars, standardize_punctuation, lemmatize, remove_stop_words, asciify_emojis, expand_contractions, remove_emojis, asciify)

python main.py ls -m fasttext -r grid_v1 -t 60 \
-p min_num_tokens min_num_chars standardize_punctuation lemmatize remove_stop_words asciify_emojis expand_contractions remove_emojis asciify