Plot

tt.plot contains lightweight plotting helpers used for experiments. They are convenience utilities rather than the core traceTorch API.

tracetorch.plot.spike_train(list_of_tensors, spacing: float = 1.0, linelength: float = 0.8, linewidth: float = 0.5, title: str = 'Spike Train Raster', use_imshow: bool = True)

Plot a spike train or signed activity sequence.

Parameters:

• list_of_tensors (Sequence[torch.Tensor]) – sequence of tensors, one per timestep. Each tensor is flattened as neuron/activity values.

• spacing (float, default=1.0) – vertical spacing for event-plot mode.

• linelength (float, default=0.8) – line length for event-plot mode.

• linewidth (float, default=0.5) – line width for event-plot mode.

• title (str, default="Spike Train Raster") – plot title.

• use_imshow (bool, default=True) – if True, draw a signed heatmap. If False, draw an event plot of nonzero entries.

Notes

This helper is intended for quick experiment visualization. It calls plt.show() and does not return the figure.

tracetorch.plot.line_graph(list_of_values, title: str, label=None) → None

Plot a simple line graph from scalar values or tensors.

Parameters:

• list_of_values – sequence of Python scalars or tensors. Tensor values are stacked over time and each flattened element is plotted as its own line.

• title (str) – plot title.

• label (Sequence, optional) – labels for tensor-valued lines.

Notes

This helper is intended for quick experiment visualization. It calls plt.show() and does not return the figure.

tracetorch.plot.render_image(tensor: torch.Tensor, title: str = None, name: str = '', save: bool = False)

Render a batch of grayscale or RGB images with Matplotlib.

Parameters:

• tensor (torch.Tensor) – image batch of shape [B, C, H, W]. C must be 1 or 3.

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

• name (str, default="") – filename stem used when save=True.

• save (bool, default=False) – if True, save the figure to media/{name}.png.

Notes

This helper is intended for quick experiment visualization. It calls plt.show() and does not return the figure.

tracetorch.plot.distributions(layers: List[torch.Tensor], title: str, *, n_grid: int = 1024, n_percentiles: int = 100, bandwidths: List[float | None] | None = None, show_percentile_lines: bool = False, compute_metrics: bool = True, max_kde_samples: int = 20000) → Dict[str, Any]

Estimate and plot value distributions for tensors.

Parameters:

• layers (List[torch.Tensor]) – tensors to flatten and compare.

• title (str) – plot title.

• n_grid (int, default=1024) – number of KDE evaluation points.

• n_percentiles (int, default=100) – number of percentile buckets.

• bandwidths (List[Optional[float]], optional) – per-layer KDE bandwidths. None entries use Silverman’s rule.

• show_percentile_lines (bool, default=False) – if True, show selected percentile markers.

• compute_metrics (bool, default=True) – if True, compute pairwise KDE L1 distance and Wasserstein distance when SciPy is available.

• max_kde_samples (int, default=20000) – maximum samples per tensor used for KDE evaluation.

Returns:

plotting data including grid, kdes, percentiles, stats, fig_ax, and optional metrics.

Return type:

Dict[str, Any]

Notes

This helper is intended for exploratory analysis and calls plt.show().

class tracetorch.plot.MeasurementManager(title: str, decay: list = [0.0, 0.9, 0.99, 0.999])

Track and plot exponentially smoothed measurements.

MeasurementManager stores raw scalar measurements and several exponential moving averages with different decays. It is useful for quick experiment tracking of losses, accuracies, or other scalar diagnostics.

Parameters:

• title (str) – default plot title.

• decay (list, default=[0., 0.9, 0.99, 0.999]) – EMA decay values to track.

append(value)

Append a scalar value and update all EMA traces.

plot(title: str = None)

Plot the stored EMA traces with tt.plot.line_graph.