fusionlab.params.LearnableC

class fusionlab.params.LearnableC

Bases: _BaseC

Indicates that the PINN’s physical coefficient \(C\) should be learned (trainable). We actually learn \(\log(C)\) to ensure \(C > 0\). The user supplies an initial_value, and the model initializes:

Trainable \(C\).

In TF mode we keep \(\log C\) as a tf.Variable, ensuring \(C>0\).

In NumPy mode the coefficient cannot be trained, so it degrades gracefully to a fixed float.

\[\log C \;=\; \log(\text{initial\_value}).\]

Parameters:

initial_value (float) – Strictly positive initial \(C\).

Variables:

initial_value (float) – The positive starting value for \(C\). Must be strictly positive.

Examples

>>> from fusionlab.params import LearnableC
>>> # Learn C, starting from C = 0.01
>>> pinn_coeff = LearnableC(initial_value=0.01)
>>> # Learn C, starting from C = 0.001
>>> pinn_coeff_small = LearnableC(initial_value=0.001)

__init__(initial_value=0.01, **kwargs)

Parameters:

initial_value (float)

Methods

__init__([initial_value])
from_config(cfg)
get_config()

Attributes

trainable

overridden by concrete classes

__init__(initial_value=0.01, **kwargs)

Parameters:

initial_value (float)