Distribution of Colours¶
This notebook plots the distribution of most excitatory and most inhibitory colours for spectrally opponent cells at different depths in our model. It corresponds to Figure 4 from the paper.
Note: The easiest way to use this is as a colab notebook, which allows you to dive in with no setup.
Load Dependencies - Colab Only¶
In [ ]:
from os.path import exists
if not exists('opponency.zip'):
!wget -O opponency.zip https://github.com/ecs-vlc/opponency/archive/master.zip
!unzip -qq opponency.zip
!mv opponency-master/* ./
!rm -r opponency-master
Generate Plots¶
In [1]:
%matplotlib inline
import matplotlib.pyplot as plt
from matplotlib import rc
import matplotlib.font_manager
rc('font',**{'family':'serif','serif':['Computer Modern Roman'],'size':13})
rc('text', usetex=True)
# plt.rcParams.update({'font.size': 18})
import pandas as pd
import numpy as np
from scipy import stats
from matplotlib import cm
from matplotlib import transforms
import matplotlib as mpl
from scipy.stats import gaussian_kde
titles = {'retina_relu2': 'Retina 2', 'ventral_relu0': 'Ventral 1',
'ventral_relu1': 'Ventral 2'}
def plot(ax, frame, layer, n_bn, d_vvs, label, title=False):
frame = frame[frame['layer'] == layer].copy()
frame['max_params'] = (frame['max_params'] * 2 * np.pi) / 360
frame['min_params'] = (frame['min_params'] * 2 * np.pi) / 360
exs = []
ins = []
main_frame = frame
for rep in range(9):
frame = main_frame[main_frame['rep'] == rep]
frame = pd.concat([frame[frame['d_vvs'] == d] for d in d_vvs])
frame = pd.concat([frame[frame['n_bn'] == n] for n in n_bn])
exc = np.histogram(frame['max_params'], bins=45, range=(0,2 * np.pi), density=True)
inh = np.histogram(frame['min_params'], bins=45, range=(0,2 * np.pi), density=True)
def get_data(x):
xs = list(x[1])
ys = list(x[0])
ys.append(ys[0])
# if(m == 0):
# return xs, ys
# ys = [(ys[i] / m) for i in range(len(ys))]
return xs, ys
exc1, exc2 = get_data(exc)
inh1, inh2 = get_data(inh)
exs.append(exc2)
ins.append(inh2)
# Define colormap normalization for 0 to 2*pi
norm = mpl.colors.Normalize(0, 2*np.pi)
# Plot a color mesh on the polar plot
# with the color set by the angle
n = 2000 #the number of secants for the mesh
t = np.linspace(0,2*np.pi,n) #theta values
r = np.linspace(0.85,1,2) #radius values change 0.6 to 0 for full circle
rg, tg = np.meshgrid(r,t) #create a r,theta meshgrid
c = tg #define color values as theta value
im = ax.pcolormesh(t, r, c.T, norm=norm, cmap='hsv') #plot the colormesh on axis with colormap
ax.set_yticklabels([])
ax.set_xticklabels([])
ax.spines['polar'].set_visible(False) #turn off the axis spine.
exs = np.array(exs)
ins = np.array(ins)
exs = (exs / np.max(np.mean(exs, axis=0) + np.std(exs, axis=0))) * 0.8
ins = (ins / np.max(np.mean(ins, axis=0) + np.std(ins, axis=0))) * 0.8
# exs = exs * 0.8
# ins = ins * 0.8
ex_mean = np.mean(exs, axis=0)
in_mean = np.mean(ins, axis=0)
ax.plot(exc1, ex_mean, linestyle=':', label='Excitation')
ax.plot(inh1, in_mean, linestyle='--', label='Inhibition')
ex_std = np.std(exs, axis=0)
in_std = np.std(ins, axis=0)
ax.fill_between(
exc1,
ex_mean + ex_std,
ex_mean - ex_std,
alpha=0.1
)
ax.fill_between(
inh1,
in_mean + in_std,
in_mean - in_std,
alpha=0.1
)
if title:
ax.set_title(titles[layer], pad=15)
if layer == 'retina_relu2':
ax.set_ylabel(label, labelpad=25, fontsize='large')
# if legend and layer == 'ventral_relu0':
# ax.legend(bbox_to_anchor=(0.5, -0.05), loc='upper center', ncol=2, frameon=False, columnspacing=1)
In [2]:
layers = ['retina_relu2', 'ventral_relu0', 'ventral_relu1']
params = [
('RGB: Random', [1, 2, 4, 8, 16, 32], [0, 1, 2, 3, 4], pd.read_pickle('statistics/devalois-random.pd')),
('RGB: Trained', [1, 2, 4, 8, 16, 32], [0, 1, 2, 3, 4], pd.read_pickle('statistics/devalois.pd'))
]
fig, axs = plt.subplots(len(params), 3, subplot_kw=dict(polar=True), gridspec_kw={'hspace': 0.1, 'wspace': 0.1})
fig.set_size_inches(9, 2.5 * len(params))
for i, layer in enumerate(layers):
for j, param in enumerate(params):
label, n_bn, d_vvs, frame = param
try:
ax = axs[j, i]
except IndexError:
ax = axs[i]
plot(ax, frame, layer, n_bn, d_vvs, label, j==0)
handles, labels = ax.get_legend_handles_labels()
fig.legend(handles, labels, bbox_to_anchor=(0.44, 0.1), loc='upper center', ncol=2, frameon=False)
plt.savefig('figures/excitatory_colours_rgb.pdf', bbox_inches='tight')
findfont: Font family ['serif'] not found. Falling back to DejaVu Sans.
In [3]:
layers = ['retina_relu2', 'ventral_relu0', 'ventral_relu1']
params = [
('Rotated (90$^\circ$)', [1, 2, 4, 8, 16, 32], [0, 1, 2, 3, 4], pd.read_pickle('statistics/devalois-distort.pd'))
]
fig, axs = plt.subplots(len(params), 3, subplot_kw=dict(polar=True), gridspec_kw={'hspace': 0.1, 'wspace': 0.1})
fig.set_size_inches(9, 2.5 * len(params))
for i, layer in enumerate(layers):
for j, param in enumerate(params):
label, n_bn, d_vvs, frame = param
try:
ax = axs[j, i]
except IndexError:
ax = axs[i]
plot(ax, frame, layer, n_bn, d_vvs, label, j==0)
handles, labels = ax.get_legend_handles_labels()
fig.legend(handles, labels, bbox_to_anchor=(0.44, 0.15), loc='upper center', ncol=2, frameon=False)
plt.savefig('figures/excitatory_colours_distort.pdf', bbox_inches='tight')
In [5]:
layers = ['retina_relu2', 'ventral_relu0', 'ventral_relu1']
params = [
('LAB: Random', [1, 2, 4, 8, 16, 32], [0, 1, 2, 3, 4], pd.read_pickle('statistics/devalois-random-lab.pd')),
('LAB: Trained', [1, 2, 4, 8, 16, 32], [0, 1, 2, 3, 4], pd.read_pickle('statistics/devalois-lab.pd')),
# ('Rotated (90$^\circ$)', [1, 2, 4, 8, 16, 32], [0, 1, 2, 3, 4], pd.read_pickle('statistics/devalois-distort.pd'))
]
fig, axs = plt.subplots(len(params), 3, subplot_kw=dict(polar=True), gridspec_kw={'hspace': 0.1, 'wspace': 0.1})
fig.set_size_inches(9, 2.5 * len(params))
for i, layer in enumerate(layers):
for j, param in enumerate(params):
label, n_bn, d_vvs, frame = param
try:
ax = axs[j, i]
except IndexError:
ax = axs[i]
plot(ax, frame, layer, n_bn, d_vvs, label, j==0)
handles, labels = ax.get_legend_handles_labels()
fig.legend(handles, labels, bbox_to_anchor=(0.44, 0.1), loc='upper center', ncol=2, frameon=False)
plt.savefig('figures/excitatory_colours_lab.pdf', bbox_inches='tight')
In [ ]: