MetaDict is designed to behave exactly like a dict while enabling (nested) attribute-style key access/assignment with IDE autocompletion support.

Many libraries claim to do the same, but fail in different ways (see Competitors).

Installation

$ pip install metadict

Features

• Attribute-style key access and assignment (dot notation) with IDE autocompletion support

  from metadict import MetaDict
  
  cfg = MetaDict()
  cfg.optimizer = 'Adam'
  print(cfg.optimizer)
  >> Adam
  

  
• Nested key assignment similar to defaultdict from collections

  cfg = MetaDict(nested_assignment=True)
  cfg.model.type = 'Transformer' 
  print(cfg.model.type)
  >> Transformer
  
  # or restrict nested assignment via context manager
  cfg = MetaDict()
  with cfg.enabling_nested_assignment() as cfg:
      cfg.model.type = 'Transformer'
  cfg.new_model.type = 'MLP'
  >> AttributeError: 'MetaDict' object has no attribute 'new_model'
  

• Is a dict

  dict_config = {'model': 'Transformer',
                 'optimizer': 'Adam'}    
  cfg = MetaDict(dict_config)
  print(isinstance(cfg, dict))
  >> True
  print(cfg == dict_config)
  >> True
  

• Inbuilt json support

  import json
      
  cfg = MetaDict({'model': 'Transformer'})
  print(json.loads(json.dumps(cfg)))
  >> {'model': 'Transformer'}
  

• Recursive conversion to dict

  cfg = MetaDict({'models': [{'name': 'Transformer'}, {'name': 'MLP'}]})
  print(cfg.models[0].name)
  >> Transformer
  
  cfg_dict = cfg.to_dict()
  print(type(cfg_dict['models'][0]))
  >> <class 'dict'>
  
  # Note: Appending a `dict` to a list within a `MetaDict` does not convert the `dict`.
  # MetaDict does not overwrite `list` so intercepting `append`. `extend`, etc. is currently not possible.
  # Simply wrap the appended or extended `dict` as a `MetaDict`.
  cfg.models.append({'name': 'RNN'})
  print(isinstance(cfg.models[-1], MetaDict))
  >> False
  
  cfg.models.append(MetaDict({'name': 'RNN'}))
  print(isinstance(cfg.models[-1], MetaDict))
  >> True
  

• No namespace conflicts with inbuilt methods like items(), update(), etc.

  cfg = MetaDict()
  # Key 'items' is assigned as in a normal dict, but a UserWarning is raised
  cfg.items = [1, 2, 3]
  >> UserWarning: 'MetaDict' object uses 'items' internally. 'items' can only be accessed via `obj['items']`.
  print(cfg)
  >> {'items': [1, 2, 3]}
  print(cfg['items'])
  >> [1, 2, 3]
  
  # But the items method is not overwritten!
  print(cfg.items)
  >> <bound method Mapping.items of {'items': [1, 2, 3]}>
  print(list(cfg.items()))
  >> [('items', [1, 2, 3])]
  

• References are preserved

  params = [1, 2, 3]    
  cfg = MetaDict({'params': params})
  print(cfg.params is params)
  >> True
  
  model_dict = {'params': params}
  cfg = MetaDict(model=model_dict)
  print(cfg.model.params is params)
  >> True
  
  # Note: dicts are recursively converted to MetaDicts, thus...
  print(cfg.model is model_dict)
  >> False
  print(cfg.model == model_dict)
  >> True
  

Documentation

Check the Test Cases for a complete overview of all MetaDict features.

Competitors

• Addict
  • No key autocompletion in IDE
  • Nested key assignment cannot be turned off
  • Newly assigned dict objects are not converted to support attribute-style key access
  • Shadows inbuilt type Dict
• Prodict
  • No key autocompletion in IDE without defining a static schema (similar to dataclass)
  • No recursive conversion of dict objects when embedded in list or other inbuilt iterables
• AttrDict
  • No key autocompletion in IDE
  • Converts list objects to tuple behind the scenes
• Munch
  • Inbuilt methods like items(), update(), etc. can be overwritten with obj.items = [1, 2, 3]
  • No recursive conversion of dict objects when embedded in list or other inbuilt iterables
• EasyDict
  • Only strings are valid keys, but dict accepts all hashable objects as keys
  • Inbuilt methods like items(), update(), etc. can be overwritten with obj.items = [1, 2, 3]
  • Inbuilt methods don't behave as expected: obj.pop('unknown_key', None) raises an AttributeError

Citation

@article{metadict,
  title = {MetaDict - Enabling dot notation and IDE autocompletion},
  author = {Hillebrand, Lars},
  year = {2022},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/LarsHill/metadict}},
}