structured_inhibition_spatial_network_model/scripts/spatial_network/analyse_hdi_optimization_orientation_map.py

import numpy as np
from brian2.units import *
from pypet import Environment
import matplotlib.pyplot as plt

from scripts.interneuron_placement import create_grid_of_excitatory_neurons, \
    create_interneuron_sheet_entropy_max_orientation, get_excitatory_neurons_in_inhibitory_axonal_clouds, Pickle, \
    get_position_mesh
from scripts.ring_network.head_direction import ex_in_network
from scripts.spatial_network.head_direction_index_over_noise_scale import excitatory_eqs, excitatory_params, \
    lif_interneuron_eqs, lif_interneuron_params, lif_interneuron_options, ei_synapse_model, ei_synapse_on_pre, \
    ei_synapse_param, ie_synapse_model, ie_synapse_on_pre, ie_synapse_param, get_synaptic_weights, \
    create_head_direction_input, calculate_rates
from scripts.spatial_network.run_entropy_maximisation_orientation_map import get_orientation_map

DATA_FOLDER = "../../data/"
LOG_FOLDER = "../../logs/"

TRAJ_NAME = "hdi_maximisation_by_differential_evolution"

def plot_and_save_sheet(ex_positions, ex_tunings, interneuron_positions, interneuron_orientations, mean_hdi, a, b, label):
    axonal_clouds = [Pickle(x, y, a, b, phi)
                             for x, y, phi in zip(*zip(*interneuron_positions), interneuron_orientations)]

    X, Y = get_position_mesh(ex_positions)

    n_ex = int(np.sqrt(len(ex_positions)))
    head_dir_preference = np.array(ex_tunings).reshape((n_ex, n_ex))

    fig = plt.figure()
    ax = fig.add_subplot(111)
    plt.set_cmap('twilight')
    c = ax.pcolor(X, Y, head_dir_preference.T, vmin=-np.pi, vmax=np.pi)
    fig.colorbar(c, ax=ax, label="Tuning")
    if axonal_clouds is not None:
        for i, p in enumerate(axonal_clouds):
            ell = p.get_ellipse()
            ax.add_artist(ell)

    plt.title('Mean HDI: {}'.format(mean_hdi))

    plt.savefig(DATA_FOLDER + 'optimization_figure_' + label + '.png')

def optimize_interneuron_orientation_by_hdi(traj):
    sheet_size = traj.orientation_map.sheet_size

    N_E = traj.network.N_E
    N_I = traj.network.N_I

    orientation_map = get_orientation_map(traj.orientation_map.correlation_length, traj.orientation_map.seed,
                                          sheet_size, N_E)
    ex_positions, ex_tunings = create_grid_of_excitatory_neurons(sheet_size,
                                                                 sheet_size,
                                                                 int(np.sqrt(N_E)), orientation_map)

    inhibitory_axon_long_axis = traj.morphology.long_axis
    inhibitory_axon_short_axis = traj.morphology.short_axis

    # Einmal die Entropy Maximierung

    inhibitory_axonal_clouds, _ = create_interneuron_sheet_entropy_max_orientation(
        ex_positions, ex_tunings, N_I, inhibitory_axon_long_axis,
        inhibitory_axon_short_axis, sheet_size,
        sheet_size, trial_orientations=30)

    initial_interneuron_orientations = [p.phi for p in inhibitory_axonal_clouds]
    interneuron_positions = [(p.x, p.y) for p in inhibitory_axonal_clouds]
    interneuron_orientation_bounds = [(-np.pi,np.pi) for x in interneuron_positions]

    mean_hdi = 1. - spatial_network_hdi_by_orientations(initial_interneuron_orientations, interneuron_positions, ex_positions, ex_tunings, traj)
    plot_and_save_sheet(ex_positions, ex_tunings, interneuron_positions, initial_interneuron_orientations, mean_hdi, inhibitory_axon_long_axis, inhibitory_axon_short_axis, 'before')


    # Und dann die optimierte Variante

    axon_cloud_save_array = np.load(DATA_FOLDER + 'current_cloud_save_array.npy')

    interneuron_positions = [(p[0], p[1]) for p in axon_cloud_save_array]
    initial_interneuron_orientations = [p[2] for p in axon_cloud_save_array]

    mean_hdi = 1. - spatial_network_hdi_by_orientations(initial_interneuron_orientations, interneuron_positions, ex_positions, ex_tunings, traj)
    plot_and_save_sheet(ex_positions, ex_tunings, interneuron_positions, initial_interneuron_orientations, mean_hdi, inhibitory_axon_long_axis, inhibitory_axon_short_axis, 'before')

    plt.show()

def spatial_network_hdi_by_orientations(interneuron_orientations, interneuron_positions, ex_positions, ex_tunings, traj):
    N_E = traj.network.N_E
    N_I = traj.network.N_I

    inhibitory_axon_long_axis = traj.morphology.long_axis
    inhibitory_axon_short_axis = traj.morphology.short_axis

    inhibitory_axonal_clouds = [Pickle(x, y, inhibitory_axon_long_axis, inhibitory_axon_short_axis, phi)
                                for x, y, phi in zip(*zip(*interneuron_positions), interneuron_orientations)]

    ie_connections = get_excitatory_neurons_in_inhibitory_axonal_clouds(ex_positions, inhibitory_axonal_clouds)

    inhibitory_synapse_strength = traj.synapse.inhibitory * nS
    excitatory_synapse_strength = traj.synapse.excitatory * mV


    ex_in_weights, in_ex_weights = get_synaptic_weights(N_E, N_I, ie_connections, excitatory_synapse_strength,
                                                        inhibitory_synapse_strength)

    sharpness = 1.0 / (traj.input.width) ** 2

    directions = np.linspace(-np.pi, np.pi, traj.input.number_of_directions, endpoint=False)

    firing_rate_array = np.ndarray((traj.N_E, traj.input.number_of_directions))

    net = ex_in_network(N_E, N_I, excitatory_eqs, excitatory_params, lif_interneuron_eqs,
                        lif_interneuron_params,
                        lif_interneuron_options, ei_synapse_model, ei_synapse_on_pre,
                        ei_synapse_param,
                        ex_in_weights, ie_synapse_model, ie_synapse_on_pre,
                        ie_synapse_param, in_ex_weights, random_seed=2)
    for dir_idx, dir in enumerate(directions):
        # We recreate the network here for every dir, which slows down the simulation quite considerably. Otherwise,
        # we get a problem with saving and restoring the spike times (0s spike for neuron 0)
        input_to_excitatory_population = create_head_direction_input(traj.input.baseline * nA, ex_tunings,
                                                                     sharpness,
                                                                     traj.input.amplitude * nA, dir)

        excitatory_neurons = net["excitatory_neurons"]
        excitatory_neurons.I = input_to_excitatory_population
        net.run(traj.simulation.duration * ms)

        exc_spike_mon = net["excitatory_spike_monitor"]
        ex_spike_trains = exc_spike_mon.spike_trains()
        ex_spike_rates = calculate_rates(ex_spike_trains.values())
        for n_idx, spike_rate in enumerate(ex_spike_rates):
            firing_rate_array[n_idx, dir_idx] = spike_rate


    #TODO: Get head direction indices

    tuning_vectors = np.zeros((N_E, traj.input.number_of_directions, 2))
    for ex_id, ex_rates in enumerate(firing_rate_array):
        rate_sum = 0.
        for dir_id, dir in enumerate(directions):
            tuning_vectors[ex_id, dir_id] = np.array([np.cos(dir), np.sin(dir)]) * ex_rates[dir_id]
            rate_sum += ex_rates[dir_id]
        if rate_sum != 0.:
            tuning_vectors[ex_id, :, :] /= rate_sum
    tuning_vectors_summed = np.sum(tuning_vectors, axis=1)

    head_direction_indices = np.array([np.linalg.norm(v) for v in tuning_vectors_summed])

    mean_hdi = np.mean(head_direction_indices)

    return 1. - mean_hdi

if __name__ == "__main__":
    print(np.linspace(0.0, 400.0, 30).tolist()[14])
    env = Environment(trajectory=TRAJ_NAME,
                      comment="Compare the head direction tuning for circular and ellipsoid interneuron morphology, "
                              "when tuning orientations to maximise entropy of connected excitatory tunings.",
                      multiproc=True, filename=DATA_FOLDER, ncores=3, overwrite_file=True, log_folder=LOG_FOLDER)

    traj = env.trajectory

    traj.f_add_parameter_group("orientation_map")

    traj.f_add_parameter("orientation_map.correlation_length", np.linspace(0.0, 400.0, 30).tolist()[14],
                         comment="Correlation length of orientations in um")
    traj.f_add_parameter("orientation_map.seed", 1, comment="Random seed for map generation.")
    traj.f_add_parameter("orientation_map.sheet_size", 450, comment="Sheet size in um")

    traj.f_add_parameter_group("network")
    traj.f_add_parameter("network.N_E", 900, comment="Number of excitatory neurons")
    traj.f_add_parameter("network.N_I", 100, comment="Number of inhibitory neurons")

    traj.f_add_parameter_group("synapse")
    traj.f_add_parameter("synapse.inhibitory", 30, "Strength of conductance-based inhibitory synapse in nS.")
    traj.f_add_parameter("synapse.excitatory", 1, "Strength of conductance-based inhibitory synapse in mV.")

    traj.f_add_parameter_group("input")
    traj.f_add_parameter("input.width", np.pi / 3.0, comment="Standard deviation of incoming head direction input.")
    traj.f_add_parameter("input.baseline", 0.2, comment="Head direction input baseline")
    traj.f_add_parameter("input.amplitude", 0.5, comment="Head direction input amplitude")
    traj.f_add_parameter("input.number_of_directions", 12, comment="Number of probed directions")

    traj.f_add_parameter_group("morphology")
    traj.f_add_parameter("morphology.long_axis", 100.0, comment="Long axis of axon ellipsoid")
    traj.f_add_parameter("morphology.short_axis", 25.0, comment="Short axis of axon ellipsoid")

    traj.f_add_parameter_group("simulation")
    traj.f_add_parameter("simulation.entropy_maximisation.steps", 30, comment="Steps for entropy maximisation")
    traj.f_add_parameter("simulation.dt", 0.1, comment="Network simulation time step in ms")
    traj.f_add_parameter("simulation.duration", 1000, comment="Network simulation duration in ms")

    optimize_interneuron_orientation_by_hdi(traj)