def feat_map_ica(layer_name,
subset_file='subset_cutoff_200_images.txt',
map_index=0,
n_plots=25):
files = get_feature_files(layer_name, subset_file)
maps = []
for idx, file in enumerate(files):
mat = np.load(file)
map_shape = mat[:, :, map_index].shape
maps.append(mat[:, :, map_index].flatten())
maps = np.stack(maps, axis=0)
ica = FastICA()
ica.fit(maps)
cols = int(np.ceil(np.sqrt(n_plots)))
rows = int(np.ceil(n_plots // cols))
fig, ax_list = plt.subplots(ncols=cols, nrows=rows)
ax_list = ax_list.flatten()
for idx, ax in enumerate(ax_list):
if idx >= n_plots:
ax.axis('off')
else:
ax.matshow(np.reshape(ica.components_[idx, :], map_shape),
interpolation='none')
ax.get_xaxis().set_visible(False)
ax.get_yaxis().set_visible(False)
plt.savefig('./plots/ica_' + layer_name + '_map_' + str(map_index) +
'.png',
format='png',
dpi=1500)
plt.close()
def inverse_avg_covariance_matrix(layer_name,
subset_file='subset_cutoff_200_images.txt',
log=False):
files = get_feature_files(layer_name, subset_file)
avg_cov = None
for idx, file in enumerate(files):
if idx % 10 == 0:
print('processing file ' + str(idx) + ' / ' + str(len(files)))
mat = np.load(file)
if avg_cov is None:
avg_cov = covariance_matrix(mat)
else:
avg_cov += covariance_matrix(mat)
avg_cov /= len(files)
inv_cov = np.linalg.pinv(avg_cov)
inv_cov[inv_cov == np.inf] = inv_cov[inv_cov != np.inf].max()
if log:
inv_cov[inv_cov == -np.inf] = inv_cov[inv_cov != -np.inf].min()
inv_cov += inv_cov.min() + 0.0000000001
norm = LogNorm(vmin=inv_cov.min(), vmax=inv_cov.max())
plt.matshow(inv_cov, norm=norm, interpolation='none')
plt.savefig('./plots/inv_avg_cov_' + layer_name + '_log.png',
format='png',
dpi=1500)
else:
inv_cov[inv_cov == -np.inf] = inv_cov[inv_cov != -np.inf].min()
plt.matshow(inv_cov, interpolation='none')
plt.savefig('./plots/inv_avg_cov_' + layer_name + '_lin.png',
format='png',
dpi=1500)
def visualize_feature_map(layer_name,
subset_file='subset_cutoff_200_images.txt',
image_index=0,
max_n=25,
highest_act=True):
file = get_feature_files(layer_name, subset_file)[image_index]
feat_map = np.load(file)
feat_map_vis(feat_map, max_n, highest_act)
def gini_index_hist(layer_name, subset_file='subset_cutoff_200_images.txt'):
files = get_feature_files(layer_name, subset_file)
ginis = []
for idx, file in enumerate(files):
if idx % 10 == 0:
print('processing file ' + str(idx) + ' / ' + str(len(files)))
mat = np.load(file)
ginis.append(gini_index(mat))
plt.hist(ginis, 50, normed=1, facecolor='green', alpha=0.75)
plt.savefig('./plots/gini_hist_' + layer_name + '.png', format='png')
def avg_gram_matrix(layer_name, subset_file='subset_cutoff_200_images.txt'):
files = get_feature_files(layer_name, subset_file)
avg_gram = None
for idx, file in enumerate(files):
if idx % 10 == 0:
print('processing file ' + str(idx) + ' / ' + str(len(files)))
mat = np.load(file)
if avg_gram is None:
avg_gram = gram_matrix(mat)
else:
avg_gram += gram_matrix(mat)
avg_gram /= len(files)
plt.matshow(avg_gram, interpolation='none')
plt.savefig('./plots/avg_gram_' + layer_name + '.png', format='png')