def seq_heatmap(self, input, annotation, seq, num):
plt.figure()
fig, ax = plt.subplots(1, 1)
ax.imshow(input, cmap=plt.cm.summer)
plt.title(f'Heatmap of the input')
ax.set_yticks(np.arange(len(annotation)))
ax.set_yticklabels(annotation)
ax.set_xticks(np.arange(len(seq)))
ax.set_xticklabels(seq)
plt.xlabel('Input seq')
plt.ylabel('Amino acid')
# Gets the path of the folder to and saves the figure in the folder if chosen in the specs.
if self.save_fig:
plot_dir = self.path + f'feature_maps/'
if num == 0:
make_folder(plot_dir)
plot_dir = plot_dir + f'feature_maps{num}/'
make_folder(plot_dir)
plt.savefig(plot_dir + f'input_sequence')
# Shows the plot if verbose is true.
if self.verbose:
plt.show()
plt.close()
def plot_max_input(self, input, iter, annotation, gen_seq):
'''
Plot a map showing the maximum input.
:param input: Data to be plotted.
:param label: The label for the input.
:param annotation: Annotation for the plots.
:param iter: How many iterations used.
:return:
'''
plt.figure()
fig, ax = plt.subplots(1, 1)
ax.imshow(input)
plt.title(f'Heatmap of the maximum input')
ax.set_yticks(np.arange(len(annotation)))
ax.set_yticklabels(annotation)
ax.set_xticks(np.arange(len(gen_seq)))
ax.set_xticklabels(gen_seq)
plt.xlabel('Input seq')
plt.ylabel('Amino acid')
# Gets the path of the folder to and saves the figure in the folder if chosen in the specs.
if self.save_fig:
plot_dir = self.path + 'max_input/'
if iter == 0:
make_folder(plot_dir)
plt.savefig(plot_dir + f'max_input_{iter}')
# Shows the plot if verbose is true.
if self.verbose:
plt.show()
plt.close()
def snp_plot(self, input, num_snp, label, annotation, input_seq):
'''
Plot changes to the score based on snp.
:param input: Data to be plotted.
:param num_sal: Numper of maps.
:param label: The label for the input.
:param annotation: Annotation for the plots.
:param input_seq: The input sequence that was used.
:return:
'''
for i in range(num_snp):
plt.figure()
fig, ax = plt.subplots(1, 1)
im = ax.imshow(input[i], cmap=plt.cm.winter)
if label[i] == 1:
plt.title(f'SNP map for input with implanted motif')
else:
plt.title(f'SNP map for input without implanted motif')
ax.set_yticks(np.arange(len(annotation)))
ax.set_yticklabels(annotation)
ax.set_xticks(np.arange(len(input_seq[i])))
ax.set_xticklabels(input_seq[i])
plt.xlabel('Input seq')
plt.ylabel('Amino acid')
fig.colorbar(im)
# Gets the path of the folder to and saves the figure in the folder if chosen in the specs.
if self.save_fig:
plot_dir = self.path + 'snp_map/'
if i == 0:
make_folder(plot_dir)
plt.savefig(plot_dir + f'SNP_implanted_{label[i]==1}_{i}')
# Shows the plot if verbose is true.
if self.verbose:
plt.show()
plt.close()
def visualize_kernel(self, input, layer, print_text):
'''
Plot heatmaps of the kernels.
:param input: Data to be plotted.
:param layer: Which layer of the model the input belongs to
:return:
'''
vmax = torch.max(input)
vmin = torch.min(input)
fig = plt.figure(constrained_layout=False, figsize=(16, 12))
if not print_text: fig.suptitle(f'Dense layer weights', size=28)
else: fig.suptitle(f'Kernels layer {layer}', size=28)
widths = np.ones(input.shape[1] + 1) * 5
widths[-1] = 1
heights = np.ones(input.shape[0])
#for i in range()
spec = gridspec.GridSpec(ncols=input.shape[1] + 1,
nrows=input.shape[0],
figure=fig,
width_ratios=widths,
height_ratios=heights)
for i in range(input.shape[0]):
for j in range(input.shape[1]):
ax = fig.add_subplot(spec[i, j])
if i == 0:
ax.xaxis.set_label_position('top')
ax.set_xticklabels([])
ax.set_xlabel(f'Filter channel {j+1}', fontsize=16)
else:
ax.xaxis.set_visible(False)
if j == 0:
ax.set_ylabel(f'Filter {i+1}', fontsize=16)
ax.set_yticklabels([])
else:
ax.yaxis.set_visible(False)
if print_text:
for y in range(input.shape[2]):
for x in range(input.shape[3]):
ax.text(x + 0,
y + 0,
f'{input[i,j,y,x]:.2f}',
horizontalalignment='center',
verticalalignment='center',
fontsize=16)
im = ax.imshow(input[i, j, :, :],
cmap=plt.cm.summer,
vmin=vmin,
vmax=vmax)
ax_cbar = fig.add_axes([0.95, 0.2, 0.02, 0.7])
cbar = plt.colorbar(im, cax=ax_cbar, orientation='vertical')
cbar.ax.tick_params(labelsize=16)
spec.tight_layout(fig, rect=[0, 0, 1, 0.95])
# Gets the path of the folder to and saves the figure in the folder if chosen in the specs.
if self.save_fig:
plot_dir = self.path + 'kernel_map/'
if layer == 1:
make_folder(plot_dir)
plt.savefig(plot_dir + f'kernels_layer{layer}')
# Shows the plot if verbose is true.
if self.verbose:
plt.show()
plt.close()
def saliency_map_plot(self, input, num_sal, label, annotation, guided,
input_seq, input_sal):
'''
Plot saliency maps.
:param input: Data to be plotted.
:param num_sal: Numper of maps.
:param label: The label for the input.
:param annotation: Annotation for the plots.
:param guided: If guided backprop is used or not.
:param input_seq: The input sequence that was used.
:return:
'''
for i in range(num_sal):
fig = plt.figure(
) #constrained_layout=True)# figsize=(10*ratio,10))
widths = [20]
heights = [22, 1]
spec = fig.add_gridspec(ncols=1,
nrows=2,
width_ratios=widths,
height_ratios=heights)
ax1 = fig.add_subplot(spec[0, 0])
ax2 = fig.add_subplot(spec[1, 0])
vmax = torch.max(input[i])
vmin = torch.min(input[i])
im1 = ax1.imshow(input[i], cmap=plt.cm.hot, vmin=vmin, vmax=vmax)
im2 = ax2.imshow(input_sal[i][None, :],
cmap=plt.cm.hot,
vmin=vmin,
vmax=vmax)
extra_title = ''
if guided: extra_title = ' with zero grad'
if label[i] == 1:
ax1.set_title(f'Saliency map for input with implanted motif' +
extra_title)
else:
ax1.set_title(
f'Saliency map for input without implanted motif' +
extra_title)
ax1.set_yticks(np.arange(len(annotation)))
ax1.set_yticklabels(annotation)
ax1.xaxis.set_visible(False)
ax2.set_xticks(np.arange(len(input_seq[i])))
ax2.set_xticklabels(input_seq[i])
ax2.yaxis.set_visible(False)
ax2.set_xlabel('Input sequence')
ax1.set_ylabel('Amino acid')
p0 = ax1.get_position().get_points().flatten()
# p1 = ax2.get_position().get_points().flatten()
ax_cbar = fig.add_axes([p0[3] - 0.08, p0[1], 0.03, p0[2] - 0.025])
plt.colorbar(im1, cax=ax_cbar, orientation='vertical')
#fig.colorbar(im1,cax=cax)
# Gets the path of the folder to and saves the figure in the folder if chosen in the specs.
if self.save_fig:
if guided: plot_dir = self.path + 'guided_saliency_map/'
else: plot_dir = self.path + 'saliency_map/'
if i == 0:
make_folder(plot_dir)
plt.savefig(plot_dir + f'saliency_implanted_{label[i]==1}_{i}')
# Shows the plot if verbose is true.
if self.verbose:
plt.show()
plt.close()