# -- import packages: ---------------------------------------------------------
import neural_diffeqs
import torch
# -- import local dependencies: -----------------------------------------------
from . import base, mix_ins
# -- set type hints: ----------------------------------------------------------
from typing import List, Literal, Optional, Union
# -- lightning model: ---------------------------------------------------------
[docs]
class LightningSDE_FixedPotential(
mix_ins.PotentialMixIn,
mix_ins.BaseForwardMixIn,
base.BaseLightningDiffEq,
):
[docs]
def __init__(
self,
latent_dim: int = 50,
name: Optional[str] = None,
mu_hidden: Union[List[int], int] = [2000, 2000],
sigma_hidden: Union[List[int], int] = [800, 800],
mu_activation: Union[str, List[str]] = "LeakyReLU",
sigma_activation: Union[str, List[str]] = "LeakyReLU",
mu_dropout: Union[float, List[float]] = 0.1,
sigma_dropout: Union[float, List[float]] = 0.1,
mu_bias: bool = True,
sigma_bias: List[bool] = True,
mu_output_bias: bool = True,
sigma_output_bias: bool = True,
mu_n_augment: int = 0,
sigma_n_augment: int = 0,
sde_type="ito",
noise_type="general",
brownian_dim=1,
coef_drift: float = 1.0,
coef_diffusion: float = 1.0,
train_lr: float = 1e-4,
train_optimizer=torch.optim.RMSprop,
train_scheduler=torch.optim.lr_scheduler.StepLR,
train_step_size: int = 10,
dt: float = 0.1,
adjoint=False,
backend="auto",
loading_existing: bool = False,
*args,
**kwargs,
) -> None:
"""
LightningSDE_FixedPotential
Parameters
----------
latent_dim : int, optional
Dimensionality of the latent space, by default 50.
name : str, optional
Name of the model, by default None.
mu_hidden : Union[List[int], int], optional
Hidden layer sizes for the drift neural network, by default [2000, 2000].
sigma_hidden : Union[List[int], int], optional
Hidden layer sizes for the diffusion neural network, by default [800, 800].
mu_activation : Union[str, List[str]], optional
Activation function(s) for the drift neural network, by default 'LeakyReLU'.
sigma_activation : Union[str, List[str]], optional
Activation function(s) for the diffusion neural network, by default 'LeakyReLU'.
mu_dropout : Union[float, List[float]], optional
Dropout rate(s) for the drift neural network, by default 0.1.
sigma_dropout : Union[float, List[float]], optional
Dropout rate(s) for the diffusion neural network, by default 0.1.
mu_bias : bool, optional
Whether to use bias in the drift neural network, by default True.
sigma_bias : List[bool], optional
Whether to use bias in the diffusion neural network, by default True.
mu_output_bias : bool, optional
Whether to use bias in the output layer of the drift neural network, by default True.
sigma_output_bias : bool, optional
Whether to use bias in the output layer of the diffusion neural network, by default True.
mu_n_augment : int, optional
Number of augmentations for the drift neural network, by default 0.
sigma_n_augment : int, optional
Number of augmentations for the diffusion neural network, by default 0.
sde_type : str, optional
Type of stochastic differential equation, by default 'ito'.
noise_type : str, optional
Type of noise, by default 'general'.
brownian_dim : int, optional
Dimensionality of the Brownian motion, by default 1.
coef_drift : float, optional
Coefficient of drift, by default 1.0.
coef_diffusion : float, optional
Coefficient of diffusion, by default 1.0.
train_lr : float, optional
Learning rate for training, by default 1e-4.
train_optimizer : torch.optim.Optimizer, optional
Optimizer for training, by default torch.optim.RMSprop.
train_scheduler : torch.optim.lr_scheduler._LRScheduler, optional
Learning rate scheduler for training, by default torch.optim.lr_scheduler.StepLR.
train_step_size : int, optional
Step size for the learning rate scheduler, by default 10.
dt : float, optional
Time step for the SDE solver, by default 0.1.
adjoint : bool, optional
Whether to use the adjoint method for the SDE solver, by default False.
backend : str, optional
Backend for the SDE solver, by default "auto".
loading_existing : bool, optional
Whether to load an existing model, by default False.
Returns
-------
None
Notes
-----
This class implements a fixed potential SDE using PyTorch Lightning.
Examples
--------
>>> model = LightningSDE_FixedPotential(latent_dim=20, dt=0.05)
>>> model.fit(data)
"""
super().__init__()
name = self._configure_name(name, loading_existing=loading_existing)
self.save_hyperparameters()
# -- torch modules: ----------------------------------------------------
self._configure_torch_modules(func=neural_diffeqs.PotentialSDE, kwargs=locals())
self._configure_lightning_model(kwargs=locals())
def __repr__(self) -> Literal['LightningSDE-FixedPotential']:
return "LightningSDE-FixedPotential"
