nemos.io.load_model

nemos.io.load_model(filename, mapping_dict=None)

Load a previously saved nemos model from a .npz file.

This will read the model parameters from the specified file and instantiate the model class with those parameters. It allows for custom mapping of attribute names to their actual objects using a mapping dictionary.

Parameters:

• filename (Union[str, Path]) – Path to the saved .npz file.

• mapping_dict (dict) – Optional dictionary to map custom attribute names to their actual objects.

Returns:

An instance of the model class with the loaded parameters.

Return type:

model

Examples

>>> import nemos as nmo
>>> # Create a GLM model with specified parameters
>>> solver_args = {"stepsize": 0.1, "maxiter": 1000, "tol": 1e-6}
>>> model = nmo.glm.GLM(
...     regularizer="Ridge",
...     regularizer_strength=0.1,
...     observation_model="Gamma",
...     solver_name="BFGS",
...     solver_kwargs=solver_args,
... )
>>> for key, value in model.get_params().items():
...     print(f"{key}: {value}")
inverse_link_function: <function one_over_x at ...>
observation_model: GammaObservations()
regularizer: Ridge()
regularizer_strength: 0.1...
solver_kwargs: {'stepsize': 0.1, 'maxiter': 1000, 'tol': 1e-06}
solver_name: BFGS
>>> # Save the model parameters to a file
>>> model.save_params("model_params.npz")
>>> # Load the model from the saved file
>>> model = nmo.load_model("model_params.npz")
>>> # Model has the same parameters before and after load
>>> for key, value in model.get_params().items():
...     print(f"{key}: {value}")
inverse_link_function: <function one_over_x at ...>
observation_model: GammaObservations()
regularizer: Ridge()
regularizer_strength: 0.1
solver_kwargs: {'maxiter': 1000, 'stepsize': 0.1, 'tol': 1e-06}
solver_name: BFGS

>>> # Loading a custom inverse link function
>>> model = nmo.glm.GLM(inverse_link_function=lambda x: x**2)
>>> model.save_params("model_params.npz")
>>> # Provide a mapping for the custom link function when loading.
>>> mapping_dict = {
...     "inverse_link_function": lambda x: x**2,
... }
>>> loaded_model = nmo.load_model("model_params.npz", mapping_dict=mapping_dict)
>>> # Now the loaded model will have the updated solver_name and solver_kwargs
>>> for key, value in loaded_model.get_params().items():
...     print(f"{key}: {value}")
inverse_link_function: <function <lambda> at ...>
observation_model: PoissonObservations()
regularizer: UnRegularized()
regularizer_strength: None
solver_kwargs: {}
solver_name: LBFGS