CANN Network Examples

This chapter covers the scripts under examples/cann/ that drive CANN models directly—1D/2D bump tracking, tuning curves, and theta-modulated visualisations. Each section lists the APIs involved, the typical output, and ideas for extending the demo.

cann1d_oscillatory_tracking.py

• Location: examples/cann/cann1d_oscillatory_tracking.py

• Goal: Run a 1D CANN under SmoothTracking1D input and export an energy-landscape animation.

• Key APIs:

  • CANN1D

  • SmoothTracking1D

  • energy_landscape_1d_animation()

• Workflow:

  1. Configure brainstate.environ.set(dt=0.1) and initialise a 512-neuron CANN.

  2. Define Iext and duration sequences, then call task.get_data() to build the stimulus buffer.

  3. Use brainstate.compile.for_loop to advance the network and collect state/history arrays.

  4. Render the animation via energy_landscape_1d_animation(); presets from PlotConfigs.energy_landscape_1d_animation still apply.

• Output: produces test_smooth_tracking_1d.gif (or the path you supply).

• Try next:

  • Swap in CANN1D_SFA to compare adaptation behaviour.

  • Feed a more complex stimulus sequence and pair the run with the tuning-curve utility below.

cann1d_tuning_curve.py

• Location: examples/cann/cann1d_tuning_curve.py

• Goal: Sample firing-rate statistics from selected neurons and plot their tuning curves.

• Key APIs:

  • CANN1D

  • SmoothTracking1D

  • tuning_curve()

• Workflow:

  1. Build a 1D CANN spanning [-π, π] with several stimulus transitions.

  2. Run the simulation, recording both firing rates and external inputs.

  3. Use PlotConfigs.tuning_curve to configure neuron indices, bin count, and styling.

  4. Call tuning_curve(); set save_path in the config to persist the plot.

• Output: interactive matplotlib figure or PNG/SVG (depending on the config).

• Try next:

  • Expand neuron_indices_to_plot or align the preferred-stimulus axis with pref_stim=cann.x.

  • Combine with predict() to compare tuning before and after learning.

cann2d_tracking.py

• Location: examples/cann/cann2d_tracking.py

• Goal: Demonstrate a 2D CANN following a sequence of targets and generate a heat-map animation.

• Key APIs:

  • CANN2D

  • SmoothTracking2D

  • energy_landscape_2d_animation()

• Workflow:

  1. Instantiate a length=100 CANN and initialise its state.

  2. Provide a list of two-dimensional Iext targets with matching duration values.

  3. Execute brainstate.compile.for_loop to step the network and collect u/r/inp tensors.

  4. Render the animation with PlotConfigs.energy_landscape_2d_animation.

• Output: CANN2D_encoding.gif by default.

• Try next:

  • Tune length or time_steps_per_second to balance resolution and runtime.

  • Reuse the 1D scripts to compare dimensionality effects.

theta_sweep_grid_cell_network.py

• Location: examples/cann/theta_sweep_grid_cell_network.py

• Goal: Run the theta-modulated direction/grid-cell pair and showcase diagnostic plots and animations.

• Key APIs:

  • DirectionCellNetwork

  • GridCellNetwork

  • theta_sweep

• Workflow:

  1. Build a OpenLoopNavigationTask to generate trajectories and theta gains.

  2. Step the networks with calculate_theta_modulation driving the oscillatory envelopes.

  3. Visualise the results using plot_population_activity_with_theta(), plot_grid_cell_manifold(), and create_theta_sweep_animation().

• Output: theta_sweep_animation.gif plus intermediate figures.

• Try next:

  • Sweep mapping_ratio or theta_strength_* to study modulation depth.

  • Feed the generated data into ThetaSweepPipeline for automated processing.

More scripts

• Tasks and Navigation Examples points to hierarchical_path_integration.py and import_external_trajectory.py, which integrate the navigation task layer.

• If you only need a quick visual check, browse the pre-rendered GIF/PNG assets in the repository root—the files beginning with test_ match the examples above.