Data package

Submodules

Data.data_losses module

This module contains all losses used for training.

class Data.data_losses.Full_MSELoss(**model_kwargs)

Bases: Module

forward(input, target)

Define the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class Data.data_losses.GradientSmoothnessLoss(model, grad_regularization=0.001)

Bases: Module

forward(x)

Define the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class Data.data_losses.Mean_MSELoss(**model_kwargs)

Bases: Module

forward(input, target)

Define the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class Data.data_losses.NMLLLoss(model, regularization_weight, **model_kwargs)

Bases: Module

Marginal Negative Log Likelihood for Gaussian Processes.

forward(input, target, isTraining=True)

Define the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class Data.data_losses.VariationalELBOLoss(model, regularization_weight, grad_regularization)

Bases: Module

forward(input, target, isTraining=True)

Define the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

Data.forecast_dataset module

This module contains all dataset wrappers used for training.

class Data.forecast_dataset.GridStationDataset(grid_ds: Dataset, station_df: DataFrame, sta_df: DataFrame, output_weather_vars: list, training_window: int)

Bases: WeatherDataset

era5_name_to_kis_name(name)

get_domain_shape()

Returns the shape of the input grid.

Returns:

torch.tensor: shape of the input grid

get_in_channels()

Returns the number of input channels.

Returns:

torch.tensor: number of input channels

get_kis_obs(idx)

get_num_train_data()

get_test_idx(idx)

get_test_loader(batch_size, shuffle=True)

get_timestamp(idx)

get_train_idx(idx)

get_train_loader(batch_size)

interpolate_era5(X: Tensor) → Tensor

Interpolates ERA5 data to target coordinates for each batch and weather variable.

Parameters:

X (Tensor): Input ERA5 data of shape [B, W, V, H]

Returns:

Tensor: Interpolated values of shape [B, W_out, M]

precompute_kis_obs()

precompute_outputs()

Precompute interpolated outputs at station locations for all timesteps.

shuffle_train_indices()

to(device)

class Data.forecast_dataset.PastNDaysForecastDataset(observations: array, training_window: int, train_ratio: float, test_ratio: float, covariances: array = None)

Bases: WeatherDataset

get_domain_shape()

Returns the shape of the input grid.

Returns:

torch.tensor: shape of the input grid

get_in_channels()

Returns the number of input channels.

Returns:

torch.tensor: number of input channels

get_test_loader(batch_size)

get_test_xy()

to(device)

class Data.forecast_dataset.WeatherDataset(train_ratio: float, test_ratio: float)

Bases: Dataset

Base class for other datasets containing weather data.

data_split()

abstract get_domain_shape() → tensor

Returns the shape of the input grid.

Returns:

torch.tensor: shape of the input grid

abstract get_in_channels() → int

Returns the number of input channels.

Returns:

torch.tensor: number of input channels

get_test_loader(batch_size)

get_test_xy()

get_train_loader(batch_size)

get_train_xy()

Data.forecast_dataset.create_GridStationDataset(training_window: int)

Data.Synthetic

This module contains the functionality for creating the synthetic dataset.

Data.Synthetic.generate_synthetic_data.compute_and_sample_values()

Computes parameters mu and Sigma for the synthetic model at three settings. Saves these parameters and samples from the probabilistic distribution.

Data.Synthetic.synthetic_data_util.compute_covariance(W, V, H, T, zeta, eps, h, c, ncycles=10)

Computes the full covariance matrix at various timesteps for the synthetic data.

Data.Synthetic.synthetic_data_util.compute_mean(alpha, beta, gamma, delta, RH, X, Z)

Computes the mean at various timesteps for the synthetic data.

Data.Synthetic.synthetic_data_util.compute_output(mu, S_full)

Computes samples at various timesteps for the synthetic data based on the means and covariance matrices.

Data.Synthetic.synthetic_data_util.get_vmins_and_vmaxs(data, p)

Data.Synthetic.synthetic_data_util.integrand(tau, t, alpha, x_vec)

Defines the integrand of the temperature solution.

Data.Synthetic.synthetic_data_util.plot_contours(data, num_time_slices=3, ncycles=10, vmins=None, vmaxs=None, savedir=None, setting=None, save_name=None, time_offset=0, model=None, closeOnSave=True, is_rel=False, saveFull=False)

General-purpose plotting function to plot synthetic data samples at various timesteps with several configurable settings.

Data.Synthetic.synthetic_data_util.sigma_t(t)

The time contribution to the covariance matrix.

Data.Synthetic.synthetic_data_util.sigma_w(d1, d2, c)

The weather type contribution to the covariance matrix.

Data.Synthetic.synthetic_data_util.sigma_x(x1, x2, eps, h)

The spatial contribution to the covariance matrix.

Data.Synthetic.synthetic_data_util.temp(z, alpha, beta, gamma, delta)

Computes the solution to the inhomogeneous forced heat equation with Gaussian IC.

Data.Synthetic.synthetic_data_util.temp_h(x_vec, t, alpha, gamma, delta)

Computes the solution to the homogeneous heat equation given the Gaussian initial temperature profile.

Data.Synthetic.synthetic_data_util.temp_i(x_vec, gamma, delta)

Provides the initial temperature profile.

Data.Synthetic.synthetic_data_util.temp_p(x_vec, t, alpha, beta)

Computes the particular temperature solution that is caused by the forcing term beta * sin(t) * delta(x).