torch_ecg.utils.make_serializable#

Recursively convert object into JSON-serializable form.

Rules#

NDArray → list
np.generic → Python scalar
dict → new dict with only serializable values
list/tuple → list with only serializable values
str/int/float/bool/None → kept
if drop_unserializable:
anything else (like Path, custom classes) → dropped (return None)

else:
fallback to str(x)

param x:: Input object to be converted.
type x:: Any
param drop_unserializable:: Whether to drop unserializable objects (return None), or convert them to string with str(x).

Added in version 0.0.32.
type drop_unserializable:: bool, default=True
param drop_paths:: If True, drop all filesystem paths (Path objects and strings that look like paths).

Added in version 0.0.32.
type drop_paths:: bool, default=False
returns:: A JSON-serializable object, or None if dropped.
rtype:: Optional[Union[list, dict, str, int, float, bool]]

Examples

>>> import numpy as np
>>> make_serializable(np.array([1, 2, 3]))
[1, 2, 3]
>>> make_serializable({"a": np.float64(3.14), "b": Path("file.txt")})
{'a': 3.14, 'b': 'file.txt'}
>>> make_serializable({"a": np.float64(3.14), "b": Path("file.txt")}, drop_paths=True)
{'a': 3.14}