brainx-visualization

Use braintools.visualize.spike_raster with flat spike-time arrays and matching neuron IDs, or per-neuron spike-train sequences. For dense time-by-neuron matrices aligned to timestamps, use raster_plot instead, and never convert flat events to a dense matrix just for plotting.

spike_raster accepts flat event arrays with matching neuron IDs or per-neuron spike trains, while raster_plot expects a dense (time, neuron) matrix aligned to a timestamp vector. Passing the wrong representation is a common failure that corrupts the displayed event structure.

Use precision-recall curves for imbalanced positive-class evaluation or when positive predictions drive the decision, since ROC can look optimistic under severe class imbalance. Always state the positive class, score direction, and prevalence, and evaluate on held-out predictions only.

Yes, helpers like interactive_spike_raster, interactive_heatmap, and dashboard_neural_activity return Plotly figures requiring Plotly installed. Export with fig.write_html for interactive artifacts, and keep a stable static companion figure for archival evidence.

This happens when each frame autoscales its axis or color limits. Fix val_min, val_max, axis limits, and colormap normalization before rendering, and set dt from the recorded interval rather than the integration step when monitoring was subsampled.

Avoid 3D when a 2D projection answers the same question without occlusion. Reserve 3D views for genuine geometric evidence like surfaces, volumes, electrodes, or three-coordinate trajectories, and route BrainCell morphology to the BrainCell topology visualization reference instead.