dymad.models.recipes_corr

Functions

enc_corr_dif_auto(self, w)

Encodes states.

enc_corr_dif_ctrl(self, w)

Encodes states and controls.

Classes

TemplateCorrAlg(model_config, data_meta[, ...])

Template class for dynamics modeling with algebraic corrections.

TemplateCorrDif(model_config, data_meta[, ...])

Template class for dynamics modeling with differential corrections.
class dymad.models.recipes_corr.TemplateCorrAlg(model_config, data_meta, dtype=None, device=None)

Bases: ComposedDynamics

Template class for dynamics modeling with algebraic corrections. Consider a base dynamics model with parameters p

  • x’ = Base_Dynamics(x, u, p)

The corrected dynamics model with residual force f is given by

  • f = Residual_Force(x, u)

  • x’ = Base_Dynamics_With_Correction(x, u, f, p)

Here the user needs to provide Base_Dynamics_With_Correction that takes the residual force as an additional input.

CONT = None
GRAPH = False
base_dynamics(x, u, f, p)

The minimal that the user must implement.

f is the residual correction term computed by self.dynamics.net, and needs to be incorporated into the base dynamics.

Return type:

Tensor

dynamics(z, w)

Processing without control inputs.

Return type:

Tensor

extra_setup()

Additional setup in derived classes. Called at end of __init__.

class dymad.models.recipes_corr.TemplateCorrDif(model_config, data_meta, dtype=None, device=None)

Bases: ComposedDynamics

Template class for dynamics modeling with differential corrections. Consider a base dynamics model with parameters p

  • x’ = Base_Dynamics(x, u, p)

Add a latent dynamics state s for correction

  • s’ = Latent_Dynamics(x, s, u)

  • f = Residual_Force(x, s, u)

Then full dynamics is

  • z = [x, Encoder(x,u)]

  • x’ = Base_Dynamics_With_Correction(x, u, f, p)

  • s’ = Latent_Dynamics(z=[x, s], u)

  • f = Residual_Force(z=[x, s], u)

  • x_hat = Decoder(z=[x, s])

Here the user needs to provide Base_Dynamics_With_Correction that takes the residual force as an additional input.

Due to the special structure, the autoencoder dimensions are fixed,

  • Encoder input: n_total_features (x or x+u)

  • Encoder output: latent_dimension (s)

  • Decoder input: n_total_state_features + latent_dimension (x+s)

  • Decoder output: n_total_state_features (x)

CONT = None
GRAPH = False
base_dynamics(x, u, f, p)

The minimal that the user must implement.

f is the residual correction term as output from the latent dynamics, and needs to be incorporated into the base dynamics.

Return type:

Tensor

diagnostic_info()

Return diagnostic information about the model.

Returns:

String with model details

Return type:

str

dynamics(z, w)

Compute the dynamics output given latent states and inputs.

Note this uses three components: features, processor, and composer.

Return type:

Tensor

extra_setup()

Additional setup in derived classes. Called at end of __init__.

dymad.models.recipes_corr.enc_corr_dif_auto(self, w)

Encodes states.

Return type:

Tensor

dymad.models.recipes_corr.enc_corr_dif_ctrl(self, w)

Encodes states and controls.

Return type:

Tensor