tempor.models.ts_ode module¶

Time series ODE model implementations.

Bases: Module

CDEFunc computes \(f_\theta for the CDE model : z_t = z_0 + \int_0^t f_\theta(z_s) dX_s\).

Parameters:¶

n_units_in : int¶: Number of input units.
n_units_hidden : int¶: Number of hidden units.
n_layers_hidden : int, optional¶: Number of hidden layers. Defaults to 1.
nonlin : Nonlin, optional¶: Nonlinearity to use in NN. Available options: Nonlin. Defaults to "relu".
dropout : float, optional¶: Dropout value. If 0, the dropout is not used. Defaults to 0.
device : Any, optional¶: PyTorch device to use. Defaults to DEVICE.

forward(t: Tensor, z: Tensor) → Tensor[source]¶: Forward pass.

training : bool¶

Bases: Module

ODEFunc computes \(f_\theta for the ODE model : z_t = z_0 + \int_0^t f_\theta(z_s) dX_s\).

Parameters:¶

n_units_hidden : int¶: Number of hidden units.
n_layers_hidden : int, optional¶: Number of hidden layers. Defaults to 1.
nonlin : Nonlin, optional¶: Nonlinearity to use in NN. Available options: Nonlin. Defaults to "relu".
dropout : float, optional¶: Dropout value. If 0, the dropout is not used. Defaults to 0.
device : Any, optional¶: PyTorch device to use. Defaults to DEVICE.

forward(t: Tensor, z: Tensor) → Tensor[source]¶: Forward pass.

training : bool¶

class tempor.models.ts_ode.ReverseGRUEncoder(n_units_in: int, n_units_latent: int, n_units_hidden: int, device: Any = device(type='cpu'))[source]¶

Bases: Module

Model (encoder and Laplace representation func). Encodes observed trajectory into latent vector.

forward(observed_data: Tensor) → Tensor[source]¶: Forward pass.

training : bool¶

class tempor.models.ts_ode.LaplaceFunc(s_dim: int, n_units_out: int, n_units_latent: int, n_units_hidden: int = 64, device: Any = device(type='cpu'))[source]¶

Bases: Module

SphereSurfaceModel : C^{b+k} -> C^{bxd} - in Riemann Sphere Coords : b dim s reconstruction terms, k is latent encoding dimension, d is output dimension.

forward(i: Tensor) → tuple[Tensor, Tensor][source]¶: Forward pass.

training : bool¶

class tempor.models.ts_ode.NeuralODE(task_type: Literal[classification] | Literal[regression], n_static_units_in: int, n_temporal_units_in: int, output_shape: list[int], n_units_hidden: int = 100, n_layers_hidden: int = 1, nonlin: Literal[none] | Literal[elu] | Literal[relu] | Literal[leaky_relu] | Literal[selu] | Literal[tanh] | Literal[sigmoid] | Literal[softmax] | Literal[gumbel_softmax] = 'relu', nonlin_out: list[tuple[Literal[none] | Literal[elu] | Literal[relu] | Literal[leaky_relu] | Literal[selu] | Literal[tanh] | Literal[sigmoid] | Literal[softmax] | Literal[gumbel_softmax], int]] | None = None, dropout: float = 0, backend: Literal[ode] | Literal[cde] | Literal[laplace] = 'cde', atol: float = 0.01, rtol: float = 0.01, interpolation: Literal[cubic] | Literal[linear] = 'cubic', ilt_reconstruction_terms: int = 33, ilt_algorithm: Literal[fourier] | Literal[dehoog] | Literal[cme] | Literal[fixed_tablot] | Literal[stehfest] = 'fourier', lr: float = 0.001, weight_decay: float = 0.001, n_iter: int = 1000, batch_size: int = 500, n_iter_print: int = 100, random_state: int = 0, patience: int = 10, n_iter_min: int = 100, clipping_value: int = 1, train_ratio: float = 0.8, device: Any = device(type='cpu'), dataloader_sampler: Sampler | None = None)[source]¶

Bases: Module

The model that computes the integral in: \(z_t = z_0 + \int_0^t f_\theta(z_s) dX_s\).

Neural ODEs are a new family of deep neural network models. Instead of specifying a discrete sequence of hidden layers, we parameterize the derivative of the hidden state using a neural network. The output of the network is computed using a blackbox differential equation solver. These are continuous-depth models that have constant memory cost, adapt their evaluation strategy to each input, and can explicitly trade numerical precision for speed.

Parameters:¶

task_type : ModelTaskType¶: The type of the problem. Available options: ModelTaskType.
n_static_units_in : int¶: Number of features in the static tensor.
n_temporal_units_in : int¶: Number of features in the temporal tensor.
output_shape : List[int]¶: Shape of the output tensor.
n_units_hidden : int, optional¶: Number of hidden units in each layer. Defaults to 100.
n_layers_hidden : int, optional¶: Number of hidden layers. Defaults to 1.
nonlin : Nonlin, optional¶: Nonlinearity to use in NN. Available options: Nonlin. Defaults to "relu".
nonlin_out : Optional[List[Tuple[Nonlin, int]]], optional¶: List of activations for the output. Example [("tanh", 1), ("softmax", 3)] - means the output layer will apply "tanh" for the first unit, and "softmax" for the following 3 units in the output. Defaults to None.
dropout : float, optional¶: Dropout value. If 0, the dropout is not used. Defaults to 0.
backend : ODEBackend, optional¶: Which solver to use: "cde", "ode", "laplace". Defaults to "cde".
atol : float, optional¶: Specific to "ode" and "cde" backends. Absolute tolerance for solution. Defaults to 1e-2.
rtol : float, optional¶: Specific to "ode" and "cde" backends. Relative tolerance for solution. Defaults to 1e-2.
interpolation : Interpolation, optional¶: Specific to "ode" and "cde" backends. "cubic" or "linear". Defaults to "cubic".
ilt_reconstruction_terms : int, optional¶: Specific to "laplace" backend. Number of ILT reconstruction terms, i.e. the number of complex \(s\) points in laplace_rep_func to reconstruct a single time point. Defaults to 33.
ilt_algorithm : ILTAlgorithm, optional¶: Specific to "laplace" backend. Inverse Laplace transform algorithm to use. Available are {fourier, dehoog, cme, fixed_tablot, stehfest}. Defaults to "fourier".
lr : float, optional¶: Learning rate for optimizer. Defaults to 1e-3.
weight_decay : float, optional¶: l2 (ridge) penalty for the weights. Defaults to 1e-3.
n_iter : int, optional¶: Maximum number of iterations. Defaults to 1000.
batch_size : int, optional¶: Batch size. Defaults to 500.
n_iter_print : int, optional¶: Number of iterations after which to print updates and check the validation loss. Defaults to 100.
random_state : int, optional¶: Random_state used. Defaults to 0.
patience : int, optional¶: Number of iterations to wait before early stopping after decrease in validation loss. Defaults to 10.
n_iter_min : int, optional¶: Minimum number of iterations to go through before starting early stopping. Defaults to 100.
clipping_value : int, optional¶: Gradients clipping value. Defaults to 1.
train_ratio : float, optional¶: Train/test split ratio. Defaults to 0.8.
device : Any, optional¶: PyTorch device to use. Defaults to DEVICE.
dataloader_sampler : Optional[sampler.Sampler], optional¶: Custom data sampler for training. Defaults to None.

forward(static_data: Tensor, temporal_data: Tensor, observation_times: Tensor) → Tensor[source]¶: Forward pass.

predict(static_data: list | ndarray | Tensor, temporal_data: list | ndarray | Tensor, observation_times: list | ndarray | Tensor) → ndarray[source]¶: Make predictions.

predict_proba(static_data: list | ndarray | Tensor, temporal_data: list | ndarray | Tensor, observation_times: list | ndarray | Tensor) → ndarray[source]¶: Predict probabilities.

score(static_data: list | ndarray | Tensor, temporal_data: list | ndarray | Tensor, observation_times: list | ndarray | Tensor, outcome: list | ndarray) → float[source]¶: Compute default score.

training : bool¶

fit(static_data: list | ndarray | Tensor, temporal_data: list | ndarray | Tensor, observation_times: list | ndarray | Tensor, outcome: list | ndarray | Tensor) → Any[source]¶: Fit (train) the model.

dataloader(static_data: Tensor, temporal_data: Tensor, observation_times: Tensor, outcome: Tensor) → tuple[DataLoader, DataLoader][source]¶: Get the train and test torch dataloaders.