make apply traverse all data container in NestedDict

chanfig/nested_dict.py

@@ -17,6 +17,7 @@
 from __future__ import annotations
 
 from functools import wraps
+from inspect import ismethod
 from os import PathLike
 from typing import TYPE_CHECKING, Any, Callable, Iterable, Iterator, Mapping, Optional, Tuple, Union
 
@@ -29,6 +30,41 @@
     from torch import dtype as TorchDtype
 
 
+def apply(obj: Any, func: Callable, *args, **kwargs) -> Any:
+    r"""
+    Apply `func` to all children of `obj`.
+
+    Note that this method is meant for non-in-place modification of `obj` and should return a new object.
+
+    Args:
+        obj: Object to apply function.
+        func: Function to be applied.
+        *args: Positional arguments to be passed to `func`.
+        **kwargs: Keyword arguments to be passed to `func`.
+
+    Returns:
+        (Any): Return value of `func`.
+
+    See Also:
+        [`apply_`][chanfig.nested_dict.apply_]: Apply a in-place operation.
+    """
+
+    if isinstance(obj, Mapping):
+        {k: apply(v, func, *args, **kwargs) for k, v in obj.items()}
+    if isinstance(obj, list):
+        [apply(v, func, *args, **kwargs) for v in obj]  # type: ignore
+    if isinstance(obj, tuple):
+        tuple(apply(v, func, *args, **kwargs) for v in obj)  # type: ignore
+    if isinstance(obj, set):
+        try:
+            set(apply(v, func, *args, **kwargs) for v in obj)  # type: ignore
+        except TypeError:
+            tuple(apply(v, func, *args, **kwargs) for v in obj)  # type: ignore
+    if isinstance(obj, NestedDict):
+        return func(*args, **kwargs) if ismethod(func) else func(obj, *args, **kwargs)
+    return obj
+
+
 class NestedDict(DefaultDict):
     r"""
     `NestedDict` further extends `DefaultDict` object by introducing a nested structure with `delimiter`.
@@ -179,18 +215,18 @@ def apply(self, func: Callable, *args, **kwargs) -> NestedDict:
             func(Callable):
 
         Examples:
-            >>> d = NestedDict()
-            >>> d.a = NestedDict()
             >>> def func(d):
             ...     d.t = 1
+            >>> d = NestedDict()
+            >>> d.a = NestedDict()
+            >>> d.b = [NestedDict(),]
+            >>> d.c = (NestedDict(),)
+            >>> d.d = {NestedDict(),}
             >>> d.apply(func).dict()
-            {'a': {'t': 1}, 't': 1}
+            {'a': {'t': 1}, 'b': [{'t': 1}], 'c': ({'t': 1},), 'd': ({'t': 1},), 't': 1}
         """
 
-        for value in self.values():
-            if isinstance(value, NestedDict):
-                value.apply(func, *args, **kwargs)
-        return func(self, *args, **kwargs) or self
+        return apply(self, func, *args, **kwargs) or self
 
     def get(self, name: Any, default: Any = Null) -> Any:
         r"""
@@ -554,7 +590,7 @@ def to(self, cls: Union[str, TorchDevice, TorchDtype]) -> Any:
             {'i': {'d': tensor(1013)}}
         """
 
-        return self.apply(lambda _: super().to(cls))
+        return self.apply(super().to, cls)
 
     def dropnull(self) -> NestedDict:
         r"""

tests/test.py

@@ -1,8 +1,9 @@
 from functools import partial
 
-from chanfig import Config, Variable
 from torch import nn
 
+from chanfig import Config, Variable
+
 
 class Model:
     def __init__(self, encoder, decoder, dropout=0.1, activation="ReLU"):