Plotting

Plotting#

NeuralFieldManifold.plottings.plot_history(history, model=None, val_loader=None, device=None, p_min=2, p_max=6)[source]#

Plot training and validation metrics over epochs.

Produces a 2×6 grid of subplots showing per-epoch loss components and accuracy for both training and validation. If a model and val_loader are supplied, the final subplot displays a confusion matrix.

Parameters:

  • history (dict) – Training history as returned by train_loop().

  • model (torch.nn.Module or None, optional) – Trained model for generating the confusion matrix.

  • val_loader (DataLoader or None, optional) – Validation data for the confusion matrix.

  • device (torch.device or None, optional) – Compute device for confusion matrix inference.

  • p_min (int, optional) – Minimum AR order for confusion matrix labels. Default is 2.

  • p_max (int, optional) – Maximum AR order for confusion matrix labels. Default is 6.

NeuralFieldManifold.plottings.plot_confusion_matrix(model, val_loader, device, p_min=2, p_max=6, ax=None)[source]#

Plot a normalised confusion matrix for AR-order predictions.

Parameters:

  • model (torch.nn.Module) – Trained model with the standard forward signature.

  • val_loader (DataLoader) – Validation data yielding (x_batch, p_batch) tuples.

  • device (torch.device) – Device on which to run inference.

  • p_min (int, optional) – Minimum AR order corresponding to class 0. Default is 2.

  • p_max (int, optional) – Maximum AR order. Default is 6.

  • ax (matplotlib.axes.Axes or None, optional) – Axes to draw on. If None, a new figure is created.

Returns:

cm – Row-normalised confusion matrix of shape (n_classes, n_classes).

Return type:

np.ndarray

NeuralFieldManifold.plottings.plot_coefficients_by_p(model, X, coeffs_true, p_true, device, p_max=6, p_min=2, title='')[source]#

Visualise predicted vs. true AR coefficients for each order class.

Creates one subplot per AR-order class showing the true coefficient trajectories (black) overlaid with the model predictions (dark red) for one sample per class.

Parameters:

  • model (torch.nn.Module) – Trained model.

  • X (array-like or torch.Tensor) – Input time series of shape (N, T).

  • coeffs_true (np.ndarray) – Ground-truth coefficients of shape (N, T, max_ar_order).

  • p_true (array-like or torch.Tensor) – 0-indexed order class labels of shape (N,).

  • device (torch.device) – Compute device.

  • p_max (int, optional) – Maximum AR order. Default is 6.

  • p_min (int, optional) – Minimum AR order. Default is 2.

  • title (str, optional) – Overall figure title.

NeuralFieldManifold.plottings.plot_tvar_sample(x, coeffs, p, W=40, P0=0.5)[source]#

Plot a single TVAR sample: coefficients, signal, and local power.

Produces a three-panel figure showing (top to bottom) the time-varying AR coefficients, the generated signal, and the sliding-window power with the constraint threshold.

Parameters:

  • x (np.ndarray) – Time series of shape (T,).

  • coeffs (np.ndarray) – AR coefficient array of shape (T, p).

  • p (int) – AR order (number of coefficient tracks to plot).

  • W (int, optional) – Sliding-window size for power computation. Default is 40.

  • P0 (float, optional) – Power constraint threshold (shown as a dashed line). Default is 0.5.

NeuralFieldManifold.pub_utils.set_pub_style()[source]#

Apply publication-quality matplotlib style defaults.

Configures global rcParams for clean, journal-ready figures with appropriate font sizes, line widths, and minimal spines.

NeuralFieldManifold.pub_utils.prettify(ax, title=None, xlabel=None, ylabel=None, add_legend=False)[source]#

Apply unified publication styling to a matplotlib Axes.

Parameters:

  • ax (matplotlib.axes.Axes) – The axes object to style.

  • title (str, optional) – Title text for the axes.

  • xlabel (str, optional) – Label for the x-axis.

  • ylabel (str, optional) – Label for the y-axis.

  • add_legend (bool, optional) – If True, add a frameless legend. Default is False.

On this page