simplify the configparser usage to get parsed attributes easily

[wyli]

Is your feature request related to a problem? Please describe.

the current usage looks tedious (https://github.com/Project-MONAI/research-contributions/blob/07da7ad36a0b218556a427f3bc5292a772c30f60/auto3dseg/algorithm_templates/segresnet2d/scripts/infer.py#L34-L50)

parser = ConfigParser()
parser.read_config(config_file_)
parser.update(pairs=_args)

data_file_base_dir = parser.get_parsed_content("data_file_base_dir")
data_list_file_path = parser.get_parsed_content("data_list_file_path")
self.fast = parser.get_parsed_content("infer")["fast"]
self.num_adjacent_slices = parser.get_parsed_content("num_adjacent_slices")
self.num_sw_batch_size = parser.get_parsed_content("num_sw_batch_size")
self.output_classes = parser.get_parsed_content("output_classes")
self.overlap_ratio = parser.get_parsed_content("overlap_ratio")
self.patch_size_valid = parser.get_parsed_content("patch_size_valid")
softmax = parser.get_parsed_content("softmax")

ckpt_name = parser.get_parsed_content("infer")["ckpt_name"]
data_list_key = parser.get_parsed_content("infer")["data_list_key"]
output_path = parser.get_parsed_content("infer")["ouptut_path"]

this is a feature request of simplifying the APIs to support more user-friendly calls
to get the parsed config as attributes directly:

cfg = ConfigParser()
cfg.read_config(config_file_)
cfg.update(pairs=_args)

data_file_base_dir = cfg.data_file_base_dir
data_list_file_path = cfg.data_list_file_path
self.fast = cfg.infer.fast
...

cc @Project-MONAI/core-reviewers

[Shadow-Devil]

Hi, I would like to work on this issue.
Is this the only place where this simplification should be done or
should I also look out for other places with similar parser.get_parsed_content() usage?

[Shadow-Devil]

E.g., here: https://github.com/Project-MONAI/research-contributions/blob/07da7ad36a0b218556a427f3bc5292a772c30f60/auto3dseg/algorithm_templates/dints/scripts/dummy_runner.py#L37-L49

[Shadow-Devil]

While searching through the files, I think I found a typo: https://github.com/Project-MONAI/research-contributions/blob/07da7ad36a0b218556a427f3bc5292a772c30f60/auto3dseg/algorithm_templates/dints/scripts/infer.py#L48
It should be output_path not ouptut_path, right?

[Shadow-Devil]

I think I got the issue wrong,
it's not about replacing the code inside research-contributions,
but to simplify the API so that the code works as described.
But I still want to work on it.

[Shadow-Devil]

My first intuition was to implement it the following way:

    def __getattr__(self, key):
        """
        Get the config by id.

        Args:
            key: id to specify the expected position. See also :py:meth:`__getitem__`.

        """
        return self[key]

Now the implementation just forwards the call to __getitem__() so we don't need to duplicate the code.
From the python documentation: Called when the default attribute access fails...

So the method won't be called if a defined method/property is tried to be accessed (e.g. config or get_parsed_content()).

One problem with this implementation is, that it only works one layer deep, so this doesn't work:

from monai.bundle.config_parser import ConfigParser
x = ConfigParser({'a':{'b':"c"}, 1:2})
x.a.b  # AttributeError: 'dict' object has no attribute 'b'

[wyli]

sounds great, I'm assigning this to you , it's ok if the first PR only works for the shallow structure...I haven't thought in detail about the nested case but we can always follow up and improve it.

(also thanks for reporting the typo!)

cc @Nic-Ma

[Nic-Ma]

Sounds good for the initial PR idea, I think we can think about the nested case together later.
But personally I feel the nested case is "must-have", otherwise, the user experience may be strange.

Thanks.