123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146 |
- import matplotlib.pyplot as plt
- import noise
- import numpy as np
- from brian2.units import *
- from scripts.spatial_maps.orientation_map import OrientationMap
- from scripts.interneuron_placement import create_interneuron_sheet_entropy_max_orientation
- from scripts.interneuron_placement import create_grid_of_excitatory_neurons, \
- create_interneuron_sheet_by_repulsive_force
- import multiprocessing
- import itertools
- trials_per_scale = 1
- N_E = 900
- N_I = 90
- sheet_x = 450 * um
- sheet_y = 450 * um
- inhibitory_axon_long_axis = 100 * um
- inhibitory_axon_short_axis = 25 * um
- number_of_excitatory_neurons_per_row = int(np.sqrt(N_E))
- '''
- Tuning Maps
- '''
- # tuning_label = "Perlin"
- tuning_label = "Orientation map"
- # optimization_label = "Repulsive"
- optimization_label = "Entropy Optimization"
- ellipse_trial_sharpening_list = []
- circle_trial_sharpening_list = []
- no_conn_trial_sharpening_list = []
- def get_fwhm_for_corr_len_and_seed(corr_len,seed):
- print(corr_len, seed)
- if tuning_label == "Perlin": # TODO: How to handle scale in Perlin
- tuning_map = lambda x, y: noise.pnoise2(x / 100.0, y / 100.0, octaves=2) * np.pi
- elif tuning_label == "Orientation map":
- map = OrientationMap(number_of_excitatory_neurons_per_row + 1, number_of_excitatory_neurons_per_row + 1,
- corr_len, sheet_x / um, sheet_y / um, seed)
- # map.improve(10)
- try:
- map.load_orientation_map()
- except:
- print(
- 'No map yet with {}x{} pixels and {} pixel correllation length and {} seed'.format(map.x_dim, map.y_dim,
- map.corr_len,
- map.rnd_seed))
- return -1,-1,-1
- tuning_map = lambda x, y: map.orientation_map(x, y)
- ex_positions, ex_tunings = create_grid_of_excitatory_neurons(sheet_x / um, sheet_y / um,
- number_of_excitatory_neurons_per_row, tuning_map)
- inhibitory_radial_axis = np.sqrt(inhibitory_axon_long_axis * inhibitory_axon_short_axis)
- if optimization_label == "Repulsive":
- inhibitory_axonal_clouds = create_interneuron_sheet_by_repulsive_force(N_I, inhibitory_axon_long_axis / um,
- inhibitory_axon_short_axis / um, sheet_x / um,
- sheet_y / um, random_seed=2, n_iterations=1000)
- inhibitory_axonal_circles = create_interneuron_sheet_by_repulsive_force(N_I, inhibitory_radial_axis / um,
- inhibitory_radial_axis / um, sheet_x / um,
- sheet_y / um, random_seed=2, n_iterations=1000)
- elif optimization_label == "Entropy Optimization":
- inhibitory_axonal_clouds, ellipse_single_trial_entropy = create_interneuron_sheet_entropy_max_orientation(
- ex_positions, ex_tunings, N_I, inhibitory_axon_long_axis / um,
- inhibitory_axon_short_axis / um, sheet_x / um,
- sheet_y / um, trial_orientations=30)
- inhibitory_axonal_circles, circle_single_trial_entropy = create_interneuron_sheet_entropy_max_orientation(
- ex_positions, ex_tunings, N_I, inhibitory_radial_axis / um,
- inhibitory_radial_axis / um, sheet_x / um,
- sheet_y / um, trial_orientations=1)
- return [ellipse_single_trial_entropy,circle_single_trial_entropy]
- corr_len_range = range(0, 451, 15)
- corr_len_range_len = len(corr_len_range)
- seed_range = range(10)
- seed_range_len = len(seed_range)
- pool_arguments = itertools.product(corr_len_range,seed_range)
- use_saved_array = False
- if not use_saved_array:
- pool = multiprocessing.Pool()
- data = pool.starmap(get_fwhm_for_corr_len_and_seed,[*pool_arguments])
- print(type(data))
- print(data)
- data_array = np.reshape(np.array(data),(corr_len_range_len,seed_range_len,2))
- data_test = []
- np.save('../../simulations/2020_02_27_entropy_over_noise_scale/data_test_save.npy', data_array)
- else:
- data_array = np.load('../../simulations/2020_02_27_entropy_over_noise_scale/data_test_save.npy')
- # ellipse_trial_sharpening_list = np.zeros((len(corr_len_range),len(seed_range)))
- # circle_trial_sharpening_list = np.zeros((len(corr_len_range),len(seed_range)))
- # no_conn_trial_sharpening_list = np.zeros((len(corr_len_range),len(seed_range)))
- print(data_array)
- no_conn_trial_sharpening_list = []
- ellipse_trial_sharpening_list = []
- circle_trial_sharpening_list = []
- for i in range(corr_len_range_len):
- ellipse_trial_sharpening_list.append(data_array[i,:,0])
- circle_trial_sharpening_list.append(data_array[i,:,1])
- print(circle_trial_sharpening_list)
- ellipse_sharpening_mean = np.array([np.mean(i) for i in ellipse_trial_sharpening_list])
- circle_sharpening_mean = np.array([np.mean(i) for i in circle_trial_sharpening_list])
- ellipse_sharpening_std_dev = np.array([np.std(i) for i in ellipse_trial_sharpening_list])
- circle_sharpening_std_dev = np.array([np.std(i) for i in circle_trial_sharpening_list])
- # print(ellipse_trial_sharpening_list)
- # print(ellipse_entropy_std_dev)
- plt.figure()
- plt.plot(corr_len_range,circle_sharpening_mean, label='Circle', marker='o',color='C1')
- plt.fill_between(corr_len_range,circle_sharpening_mean-circle_sharpening_std_dev,circle_sharpening_mean+circle_sharpening_std_dev,color='C1',alpha=0.4)
- plt.plot(corr_len_range,ellipse_sharpening_mean, label='Ellipse', marker='o',color='C2')
- plt.fill_between(corr_len_range,ellipse_sharpening_mean-ellipse_sharpening_std_dev,ellipse_sharpening_mean+ellipse_sharpening_std_dev,color='C2',alpha=0.4)
- plt.xlabel('Correlation length')
- plt.ylabel('Entropy')
- plt.legend()
- plt.show()
- # plt.figure()
- # plt.plot(corr_len_list,circle_entropy_mean, label='Circle', marker='o',color='C1')
- # plt.fill_between(corr_len_list,circle_entropy_mean-circle_entropy_std_dev,circle_entropy_mean+circle_entropy_std_dev,color='C1',alpha=0.4)
- # plt.plot(corr_len_list,ellipse_entropy_mean, label='Ellipse', marker='o',color='C2')
- # plt.fill_between(corr_len_list,ellipse_entropy_mean-ellipse_entropy_std_dev,ellipse_entropy_mean+ellipse_entropy_std_dev,color='C2',alpha=0.4)
- # plt.xlabel('Correlation length')
- # plt.ylabel('Entropy')
- # plt.legend()
- # plt.show()
|