autoBOTLib CLI

To streamline the experiments, it makes a lot of sense to directly use the autoBOTLib as a CLI tool. The library itself implements wrappers for main functions, and can be executed as follows. If you installed the package, the autobot-cli tool was also built as part of the installation. By running

autobot-cli --help

The arguments are defined as follows.

usage: autobot-cli [-h] [--time TIME] [--job_id JOB_ID] [--popsize POPSIZE]
           [--output_folder OUTPUT_FOLDER]
           [--learner_preset LEARNER_PRESET] [--hof_size HOF_SIZE]
           [--representation_type REPRESENTATION_TYPE]
           [--train_data TRAIN_DATA] [--mutation_rate MUTATION_RATE]
           [--crossover_rate CROSSOVER_RATE]
           [--predict_data PREDICT_DATA] [--load_model LOAD_MODEL]
           [--num_cpu NUM_CPU] [--upsample UPSAMPLE] [--mode MODE]

optional arguments:
  -h, --help            show this help message and exit
  --time TIME
  --job_id JOB_ID
  --popsize POPSIZE
  --output_folder OUTPUT_FOLDER
  --learner_preset LEARNER_PRESET
  --hof_size HOF_SIZE
  --representation_type REPRESENTATION_TYPE
  --train_data TRAIN_DATA
  --mutation_rate MUTATION_RATE
  --crossover_rate CROSSOVER_RATE
  --predict_data PREDICT_DATA
  --load_model LOAD_MODEL
  --num_cpu NUM_CPU
  --upsample UPSAMPLE
  --mode MODE

Hence, as a minimal example, we can consider running

autobot-cli --train_data ./data/insults/train.tsv --output_folder CLI --learner_preset mini-l1

Here, the train.tsv file needs to have two attributes; text_a - the documents field, and label, the label field. Once the run finishes, you will have a trained model with a report in the CLI folder. To make predictions on unseen data, simply

autobot-cli --mode prediction --predict_data ./data/insults/test.tsv --load_model ./CLI/autoBOTmodel.pickle --output_folder CLI

See https://github.com/skblaz/autobot/cli_example.sh for a full example.