umap
umap copied to clipboard
Recommended way to save umap instances and reliably load it
How to save umap instances and reliably load it (also in higher package versions)
I would like to be able to load umap models fitted and pickled in 0.5.3 also in >=0.5.7 to avoid creating a version dependency for these models. Is there a recommended way to save and reload model weights?
Trying to unpickle the umap object in 0.5.7 leads to the following error related to numba:
Traceback (most recent call last):
saved_data = pickle.load(f)
^^^^^^^^^^^^^^
File "../lib/python3.11/site-packages/numba/core/serialize.py", line 110, in _unpickle__CustomPickled
ctor, states = loads(serialized)
^^^^^^^^^^^^^^^^^
TypeError: code() argument 13 must be str, not int