structured_inhibition_spatial_network_model/scripts/spatial_network/orientation_map/run_orientation_map.py

import json
import os

import numpy as np
from brian2.units import *
from pypet import Environment, cartesian_product, Trajectory
from pypet.brian2.parameter import Brian2MonitorResult

from scripts.interneuron_placement import create_grid_of_excitatory_neurons, \
    create_interneuron_sheet_entropy_max_orientation, get_excitatory_neurons_in_inhibitory_axonal_clouds
from scripts.ring_network.head_direction import ex_in_network
from scripts.spatial_maps.orientation_maps.orientation_map import OrientationMap
from scripts.spatial_maps.orientation_maps.orientation_map_generator_pypet import TRAJ_NAME_ORIENTATION_MAPS
from scripts.spatial_maps.uniform_perlin_map import UniformPerlinMap
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

POLARIZED = 'ellipsoid'
CIRCULAR = 'circular'
NO_SYNAPSES = 'no conn'

def get_local_data_folder():
    data_folder = "../../../data/"
    config_file_name = ".config.json"
    if os.path.isfile(config_file_name):
        with open(config_file_name)  as config_file:
            config_dict = json.load(config_file)
            data_folder = os.path.abspath(config_dict["data"])
            print(data_folder)

    return data_folder


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

TRAJ_NAME = "full_figure_orientation_map"


def get_orientation_map(correlation_length, seed, sheet_size, N_E, data_folder=None):
    if data_folder is None:
        data_folder = DATA_FOLDER

    traj = Trajectory(filename=data_folder + TRAJ_NAME_ORIENTATION_MAPS + ".hdf5")

    traj.f_load(index=-1, load_parameters=2, load_results=2)

    available_lengths = sorted(list(set(traj.f_get("corr_len").f_get_range())))
    closest_length = available_lengths[np.argmin(np.abs(np.array(available_lengths)-correlation_length))]
    if closest_length!=correlation_length:
        print("Warning: desired correlation length {:.1f} not available. Taking {:.1f} instead".format(
            correlation_length, closest_length))
    corr_len = closest_length
    seed = seed

    map_by_params = lambda x, y: x == corr_len and y == seed

    idx_iterator = traj.f_find_idx(['corr_len', 'seed'], map_by_params)

    # TODO: Since it has only one entry, maybe iterator can be replaced
    for idx in idx_iterator:
        traj.v_idx = idx
        map_angle_grid = traj.crun.map

    number_of_excitatory_neurons_per_row = int(np.sqrt(N_E))

    map = OrientationMap(number_of_excitatory_neurons_per_row, number_of_excitatory_neurons_per_row,
                         corr_len, sheet_size, sheet_size, seed)
    map.angle_grid = map_angle_grid

    return map.tuning

def get_uniform_orientation_map(correlation_length, seed, sheet_size, N_E, data_folder=None):
    if data_folder is None:
        data_folder = DATA_FOLDER

    traj = Trajectory(filename=data_folder + TRAJ_NAME_ORIENTATION_MAPS + ".hdf5")

    traj.f_load(index=-1, load_parameters=2, load_results=2)

    available_lengths = sorted(list(set(traj.f_get("corr_len").f_get_range())))
    closest_length = available_lengths[np.argmin(np.abs(np.array(available_lengths)-correlation_length))]
    if closest_length!=correlation_length:
        print("Warning: desired correlation length {:.1f} not available. Taking {:.1f} instead".format(
            correlation_length, closest_length))
    corr_len = closest_length
    seed = seed

    map_by_params = lambda x, y: x == corr_len and y == seed

    idx_iterator = traj.f_find_idx(['corr_len', 'seed'], map_by_params)

    # TODO: Since it has only one entry, maybe iterator can be replaced
    for idx in idx_iterator:
        traj.v_idx = idx
        map_angle_grid = traj.crun.map

    number_of_excitatory_neurons_per_row = int(np.sqrt(N_E))

    map = OrientationMap(number_of_excitatory_neurons_per_row, number_of_excitatory_neurons_per_row,
                         corr_len, sheet_size, sheet_size, seed)

    # Uniformize orientation map

    nrow = number_of_excitatory_neurons_per_row
    size = sheet_size
    scale = corr_len  # TODO: Probably this needs to be a linear interpolation

    n = map_angle_grid / np.pi
    m = np.concatenate(n)
    sorted_idx = np.argsort(m)
    max_val = nrow * 2
    idx = len(m) // max_val
    for ii, val in enumerate(range(max_val)):
        m[sorted_idx[ii * idx:(ii + 1) * idx]] = val
    p_map = (m - nrow) / nrow
    map.angle_grid = p_map.reshape(nrow, -1) * np.pi
    # self.map *= np.pi

    # map.angle_grid = map_angle_grid

    return map.tuning


def spatial_network_with_entropy_maximisation(traj):
    sheet_size = traj.map.sheet_size

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

    orientation_map = get_uniform_orientation_map(traj.map.correlation_length, traj.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

    entropy_maximisation_steps = traj.simulation.entropy_maximisation.steps if inhibitory_axon_long_axis != \
                                                                               inhibitory_axon_short_axis else 1

    inhibitory_axonal_clouds, ellipse_single_trial_entropy = 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=entropy_maximisation_steps)

    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

    if inhibitory_synapse_strength != 0.0 * nS and excitatory_synapse_strength != 0.0 * mV \
            and inhibitory_axon_long_axis == inhibitory_axon_short_axis:
        traj.f_add_derived_parameter("morphology.morph_label", CIRCULAR,
                                     comment="Interneuron morphology of this run is circular")
    elif inhibitory_synapse_strength != 0.0 * nS and excitatory_synapse_strength != 0.0 * mV:
        traj.f_add_derived_parameter("morphology.morph_label", POLARIZED,
                                     comment="Interneuron morphology of this run is ellipsoid")
    else:
        traj.f_add_derived_parameter("morphology.morph_label", NO_SYNAPSES,
                                     comment="There are no interneurons")

    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 = get_input_head_directions(traj)
    for 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)
        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)
        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

        inhibitory_neurons = net["interneurons"]
        inhibitory_neurons.u_ext = traj.inh_input.baseline * mV
        inhibitory_neurons.tau = traj.interneuron.tau * ms

        net.run(traj.simulation.duration * ms)

        direction_id = 'dir{:d}'.format(idx)
        traj.f_add_result(Brian2MonitorResult, '{:s}.spikes.e'.format(direction_id), net["excitatory_spike_monitor"],
                          comment='The spiketimes of the excitatory population')
        traj.f_add_result(Brian2MonitorResult, '{:s}.spikes.i'.format(direction_id), net["inhibitory_spike_monitor"],
                          comment='The spiketimes of the inhibitory population')
    traj.f_add_result('ex_positions', np.array(ex_positions),
                      comment='The positions of the excitatory neurons on the sheet')
    traj.f_add_result('ex_tunings', np.array(ex_tunings),
                      comment='The input tunings of the excitatory neurons')

    ie_connections_save_array = np.zeros((N_I, N_E))
    for i_idx, ie_conn in enumerate(ie_connections):
        for e_idx in ie_conn:
            ie_connections_save_array[i_idx, e_idx] = 1
    traj.f_add_result('ie_adjacency', ie_connections_save_array,
                      comment='Recurrent connection adjacency matrix')

    axon_cloud_save_list = [[p.x, p.y, p.phi] for p in inhibitory_axonal_clouds]
    axon_cloud_save_array = np.array(axon_cloud_save_list)
    traj.f_add_result('inhibitory_axonal_cloud_array', axon_cloud_save_array,
                      comment='The inhibitory axonal clouds')

    return 1


def get_input_head_directions(traj):
    directions = np.linspace(-np.pi, np.pi, traj.input.number_of_directions, endpoint=False)
    return directions


def main():
    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=0, overwrite_file=True, log_folder=LOG_FOLDER)

    traj = env.trajectory

    traj.f_add_parameter_group("map")

    traj.f_add_parameter("map.correlation_length", 200.0,
                         comment="Correlation length of orientations in um")
    traj.f_add_parameter("map.seed", 1, comment="Random seed for map generation.")
    traj.f_add_parameter("map.sheet_size", 900, comment="Sheet size in um")

    traj.f_add_parameter_group("network")
    traj.f_add_parameter("network.N_E", 3600, comment="Number of excitatory neurons")
    traj.f_add_parameter("network.N_I", 400, comment="Number of inhibitory neurons")

    traj.f_add_parameter_group("interneuron")
    traj.f_add_parameter("interneuron.tau", 7., comment="Interneuron timescale in ms")

    traj.f_add_parameter_group("synapse")
    traj.f_add_parameter("synapse.inhibitory", 30.0, "Strength of conductance-based inhibitory synapse in nS.")
    traj.f_add_parameter("synapse.excitatory", 2.5, "Strength of conductance-based inhibitory synapse in mV.")

    traj.f_add_parameter_group("input")
    traj.f_add_parameter("input.width", 1. / np.sqrt(2.5), comment="Standard deviation of incoming head direction input.")
    traj.f_add_parameter("input.baseline", 0.05, comment="Head direction input baseline")
    traj.f_add_parameter("input.amplitude", 0.6, comment="Head direction input amplitude")
    traj.f_add_parameter("input.number_of_directions", 12, comment="Number of probed directions")

    traj.f_add_parameter_group("inh_input")
    traj.f_add_parameter("inh_input.baseline", -50., comment="Head direction input baseline")
    traj.f_add_parameter("inh_input.amplitude", 0., comment="Head direction input amplitude")

    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")

    correlation_length_range = np.linspace(1.0, 800.0, 12, endpoint=True).tolist()
    # correlation_length_range = [200.0]
    seed_range = range(10)
    # seed_range = [1]

    ellipsoid_parameter_exploration = {
        "morphology.long_axis": [100.0],
        "morphology.short_axis": [25.0],
        "map.correlation_length": correlation_length_range,
        "map.seed": seed_range,
        "synapse.inhibitory": [30.],
        "synapse.excitatory": [2.5]
        # "map.correlation_length": np.arange(0.0, 200.0, 50).tolist()
    }

    corresponding_circular_radius = float(np.sqrt(ellipsoid_parameter_exploration[
                                                      "morphology.long_axis"][0] * ellipsoid_parameter_exploration[
                                                      "morphology.short_axis"][0]))

    circle_parameter_exploration = {
        "morphology.long_axis": [corresponding_circular_radius],
        "morphology.short_axis": [corresponding_circular_radius],
        "map.correlation_length": ellipsoid_parameter_exploration["map.correlation_length"],
        "map.seed": ellipsoid_parameter_exploration["map.seed"],
        "synapse.inhibitory": ellipsoid_parameter_exploration["synapse.inhibitory"],
        "synapse.excitatory": ellipsoid_parameter_exploration["synapse.excitatory"]
    }

    no_conn_parameter_exploration = {
        "morphology.long_axis": [corresponding_circular_radius],
        "morphology.short_axis": [corresponding_circular_radius],
        "map.correlation_length": ellipsoid_parameter_exploration["map.correlation_length"],
        "map.seed": ellipsoid_parameter_exploration["map.seed"],
        "synapse.inhibitory": [0.],
        "synapse.excitatory": [0.]
    }

    expanded_dicts = [cartesian_product(dict) for dict in [ellipsoid_parameter_exploration,
                                                           circle_parameter_exploration,
                                                           no_conn_parameter_exploration]]
    final_dict = {}

    for key in expanded_dicts[0].keys():
        list_of_parameter_lists = [dict[key] for dict in expanded_dicts]
        final_dict[key] = sum(list_of_parameter_lists, [])

    traj.f_explore(final_dict)
    env.run(spatial_network_with_entropy_maximisation)

    env.disable_logging()

if __name__ == "__main__":
    main()