torchcell.viz.graph_recovery.GraphRecoveryVisualization

class GraphRecoveryVisualization(base_dir: str, mplstyle_path: str | None = None)

Bases: object

Visualization class for graph regularization and edge recovery metrics.

save_and_log_figure(fig: Figure, name: str, timestamp_str: str | None = None) → None

Log figure to wandb only (no disk save).

Parameters:

• fig – Matplotlib figure

• name – Wandb panel key (e.g., “val_edge_recovery_summary/recall”)

• timestamp_str – Unused, kept for API compatibility

plot_graph_info_summary(graph_info: dict[str, dict[str, float]], save_path: str | None = None) → None

Create a comprehensive summary visualization of graph statistics.

Parameters:

• graph_info – Dict mapping graph_name -> {num_edges, num_nodes, avg_degree, reg_layer, reg_head}

• save_path – Optional path to save the figure to disk

plot_edge_recovery_recall(recall_metrics: dict[str, float], num_epochs: int, timestamp_str: str | None = None, stage: str = 'val') → None

Plot recall@degree for each graph as a bar chart.

Parameters:

• recall_metrics – Dict mapping metric_key (e.g., “physical_L0_H1”) -> recall@degree value

• num_epochs – Current epoch number

• timestamp_str – Unused, kept for API compatibility

• stage – Stage name (e.g., “val”)

plot_edge_recovery_precision(precision_metrics: dict[str, dict[int, float]], k_values: list[int], num_epochs: int, timestamp_str: str | None = None, stage: str = 'val') → None

Plot precision@k for each graph as line plots.

Parameters:

• precision_metrics – Dict mapping metric_key (e.g., “physical_L0_H1”) -> {k -> precision value}

• k_values – List of k values (e.g., [8, 32, 128, 320])

• num_epochs – Current epoch number

• timestamp_str – Unused, kept for API compatibility

• stage – Stage name (e.g., “val”)

plot_edge_recovery_per_graph(recall_metrics: dict[str, float], precision_metrics: dict[str, dict[int, float]], k_values: list[int], num_epochs: int, timestamp_str: str | None = None, stage: str = 'val') → None

Create individual edge recovery plots for each graph showing both recall and precision@k.

Parameters:

• recall_metrics – Dict mapping graph_name -> recall@degree value

• precision_metrics – Dict mapping graph_name -> {k -> precision value}

• k_values – List of k values (e.g., [8, 32, 128, 320])

• num_epochs – Current epoch number

• timestamp_str – Unused, kept for API compatibility

• stage – Stage name (e.g., “val”)

plot_edge_mass_alignment(edge_mass_metrics: dict[str, float], num_epochs: int, timestamp_str: str | None = None, stage: str = 'val') → None

Plot edge-mass alignment for each graph as a bar chart.

Edge-mass alignment = fraction of total attention mass on known graph edges.

Parameters:

• edge_mass_metrics – Dict mapping metric_key (e.g., “physical_L0_H1”) -> edge_mass_fraction

• num_epochs – Current epoch number

• timestamp_str – Unused, kept for API compatibility

• stage – Stage name (e.g., “val”)