def test16_retrain_average(self):
shutil.rmtree(OUTPUT_DIR, ignore_errors=True)
args = [
"--data-dir",
DATA_BIN,
"--output-dir",
OUTPUT_DIR,
"--model-dir",
MODEL_DIR,
"--tags",
"CLI_tests",
"--lr",
"0.3",
"--epochs",
"5",
"--log-every",
"10",
"--new-classes",
"IN:GET_LOCATION,IN:GET_LOCATION_HOME,SL:POINT_ON_MAP,SL:CATEGORY_LOCATION",
"--batch-size",
"16",
"--no-lr-scheduler",
"--average-checkpoints",
"--new-model-weight",
"1",
]
args = retrain.parse_args(args)
retrain.main(args)
def test18_retrain_weight_consolidation(self):
shutil.rmtree(OUTPUT_DIR, ignore_errors=True)
args = [
"--data-dir",
DATA_BIN,
"--output-dir",
OUTPUT_DIR,
"--model-dir",
MODEL_DIR,
"--tags",
"CLI_tests",
"--lr",
"0.3",
"--epochs",
"5",
"--log-every",
"10",
"--new-classes",
"IN:GET_LOCATION,IN:GET_LOCATION_HOME,SL:POINT_ON_MAP,SL:CATEGORY_LOCATION",
"--batch-size",
"16",
"--weight-consolidation",
"0.1",
]
args = retrain.parse_args(args)
retrain.main(args)
def test15_retrain_simple_no_sched(self):
shutil.rmtree(OUTPUT_DIR, ignore_errors=True)
args = [
"--data-dir",
DATA_BIN,
"--output-dir",
OUTPUT_DIR,
"--model-dir",
MODEL_DIR,
"--tags",
"CLI_tests",
"--lr",
"0.3",
"--epochs",
"1",
"--log-every",
"10",
"--new-classes",
"IN:GET_LOCATION,IN:GET_LOCATION_HOME,SL:POINT_ON_MAP,SL:CATEGORY_LOCATION",
"--batch-size",
"16",
"--no-lr-scheduler",
]
args = retrain.parse_args(args)
retrain_simple.check_args(args)
args = retrain_simple.set_default_args(args)
retrain_simple.main(args)
def test14_retrain_limit_iters(self):
shutil.rmtree(OUTPUT_DIR, ignore_errors=True)
args = [
"--data-dir",
DATA_BIN,
"--output-dir",
OUTPUT_DIR,
"--model-dir",
MODEL_DIR,
"--tags",
"CLI_tests",
"--epochs",
"1",
"--min-steps",
"3",
"--max-steps",
"10",
"--new-classes",
"IN:GET_LOCATION,IN:GET_LOCATION_HOME,SL:POINT_ON_MAP,SL:CATEGORY_LOCATION",
"--label-smoothing",
"0.18",
]
args = retrain.parse_args(args)
retrain.main(args)
def test06_retrain_old_data001_merge(self):
shutil.rmtree(OUTPUT_DIR, ignore_errors=True)
args = [
"--data-dir",
DATA_BIN,
"--output-dir",
OUTPUT_DIR,
"--model-dir",
MODEL_DIR,
"--tags",
"CLI_tests",
"--epochs",
"1",
"--new-classes",
"IN:GET_LOCATION,IN:GET_LOCATION_HOME,SL:POINT_ON_MAP,SL:CATEGORY_LOCATION",
"--old-data-amount",
"0.01",
"--early-stopping",
"16",
"--old-data-sampling-method",
"merge_subset",
]
args = retrain.parse_args(args)
retrain.main(args)
def test05_retrain_noargs(self):
shutil.rmtree(OUTPUT_DIR, ignore_errors=True)
args = [
"--data-dir",
DATA_BIN,
"--output-dir",
OUTPUT_DIR,
"--model-dir",
MODEL_DIR,
"--tags",
"CLI_tests",
"--epochs",
f"{EPOCHS}",
"--new-classes",
"IN:GET_LOCATION,IN:GET_LOCATION_HOME,SL:POINT_ON_MAP,SL:CATEGORY_LOCATION",
]
args = retrain.parse_args(args)
retrain.main(args)
logger.info(description)
wandb_logger.log_metrics({
**final_metrics["means"],
**final_metrics["stdevs"]
})
# Compute RI and RD
class_weights = eval_dataset.get_class_frequencies(schema_tokenizer)
class_weights = {
f"cls/eval_{cls}_tree_path_f1": p
for cls, p in class_weights.items()
}
finetuning_metrics = cli_utils.evaluate_finetuning_procedure(
pretrain_metrics, final_metrics, class_weights)
wandb_logger.log_metrics(finetuning_metrics)
wandb_logger.close()
if args.clean_output:
shutil.rmtree(args.output_dir)
if __name__ == "__main__":
args = cli_retrain.parse_args()
check_args(args)
args = set_default_args(args)
main(args)