Trainer.parse_argparser does not yield sensible default for default_root_dir #1916
Description
🐛 Bug
Using Trainer.parse_argparser returns True for default_root_dir, however, a string is expected.
To Reproduce
Steps to reproduce the behavior:
>>> from pytorch_lightning import Trainer
>>> from argparse import ArgumentParser, Namespace
>>> parser = ArgumentParser(add_help=False)
>>> parser = Trainer.add_argparse_args(parent_parser=parser)
>>> args = Trainer.parse_argparser(parser)
>>> args
Namespace(accumulate_grad_batches=1, amp_level='O1', auto_lr_find=False, auto_scale_batch_size=False, auto_select_gpus=False, benchmark=False, check_val_every_n_epoch=1, checkpoint_callback=True, default_root_dir=True, deterministic=False, distributed_backend=True, early_stop_callback=False, fast_dev_run=False, gpus=<function Trainer._arg_default at 0x1219efdd0>, gradient_clip_val=0, log_gpu_memory=True, log_save_interval=100, logger=True, max_epochs=1000, max_steps=True, min_epochs=1, min_steps=True, num_nodes=1, num_processes=1, num_sanity_val_steps=2, overfit_pct=0.0, precision=32, print_nan_grads=False, process_position=0, profiler=True, progress_bar_callback=True, progress_bar_refresh_rate=1, reload_dataloaders_every_epoch=False, replace_sampler_ddp=True, resume_from_checkpoint=True, row_log_interval=10, terminate_on_nan=False, test_percent_check=1.0, tpu_cores=True, track_grad_norm=-1, train_percent_check=1.0, truncated_bptt_steps=True, val_check_interval=1.0, val_percent_check=1.0, weights_save_path=True, weights_summary='full')