train.py

import json
import shutil
import sys

from allennlp.commands import main

config_file = "train_configs/defaults.jsonnet"

# Use overrides to train on CPU
# overrides = json.dumps({"trainer": {"cuda_device": -1}})

# Use overrides to train on GPU
overrides = json.dumps({"trainer": {"cuda_device": 0}})

serialization_dir = "experiments/name_of_experiment"

# Training will fail if the serialization directory already
# has stuff in it. If you are running the same training loop
# over and over again for debugging purposes, it will.
# Hence we wipe it out in advance.
# BE VERY CAREFUL NOT TO DO THIS FOR ACTUAL TRAINING!
shutil.rmtree(serialization_dir, ignore_errors=True)

# Assemble the command into sys.argv
sys.argv = [
    "allennlp",  # command name, not used by main
    "train",
    config_file,
    "-s", serialization_dir,
    "--include-package", "dataset_readers.spider",
    "--include-package", "models.semantic_parsing.spider_parser",
    "-o", overrides,
]


import argparse
parser = argparse.ArgumentParser()
parser.add_argument('--toy', action='store_true', default=False,
        help='If set, use small data; used for fast debugging.')

main()