def plot_result(preds, labels, images):
'''
Display 5 images along with labels and predictions
Args:
preds (np array): array of predicted classes
labels (np array): array of actual classes
images (np array): array of images
'''
# set the title of the plot
title = ""
for ii in range(5):
title = title + 3 * "-" + " {}({}) ".format(preds[ii],
labels[ii]) + 3 * "-"
imshow_(torchvision.utils.make_grid(images[:5]), ax=(16, 6), title=title)
if not os.path.isdir(dir_out):
os.makedirs(dir_out)
out = {"P":P_test, "Y":y_test, "S":AP_test}
sio.savemat( dir_out+"/{}.mat".format(split), out)
# ---- Viz predictions -----
if viz_predictions:
max_classes = data.n_classes - 1
# # Output all truth/prediction pairs
plt.figure(split, figsize=(20,10))
P_test_ = np.array(P_test)/float(n_classes-1)
y_test_ = np.array(y_test)/float(n_classes-1)
for i in range(len(y_test)):
P_tmp = np.vstack([y_test_[i], P_test_[i]])
plt.subplot(n_test,1,i+1); imshow_(P_tmp, vmin=0, vmax=1)
plt.xticks([])
plt.yticks([])
acc = np.mean(y_test[i]==P_test[i])*100
plt.ylabel("{:.01f}".format(acc))
# plt.title("Acc: {:.03}%".format(100*np.mean(P_test[i]==y_test[i])))
# ---- Viz weights -----
if viz_weights and model_type is "TCN":
# Output weights at the first layer
plt.figure(2, figsize=(15,15))
ws = model.get_weights()[0]
for i in range(min(36, len(ws.T))):
plt.subplot(6,6,i+1)
# imshow_(model.get_weights()[0][i][:,:,0]+model.get_weights()[1][i])
imshow_(np.squeeze(ws[:,:,:,i]).T)
os.makedirs(dir_out)
out = {"P": P_test, "Y": y_test, "S": AP_test}
sio.savemat(dir_out + "/{}.mat".format(split), out)
# ---- Viz predictions -----
if viz_predictions:
max_classes = data.n_classes - 1
# # Output all truth/prediction pairs
plt.figure(split, figsize=(20, 10))
P_test_ = np.array(P_test) / float(n_classes - 1)
y_test_ = np.array(y_test) / float(n_classes - 1)
for i in range(len(y_test)):
P_tmp = np.vstack([y_test_[i], P_test_[i]])
plt.subplot(n_test, 1, i + 1)
imshow_(P_tmp, vmin=0, vmax=1)
plt.xticks([])
plt.yticks([])
acc = np.mean(y_test[i] == P_test[i]) * 100
plt.ylabel("{:.01f}".format(acc))
# plt.title("Acc: {:.03}%".format(100*np.mean(P_test[i]==y_test[i])))
# ---- Viz weights -----
if viz_weights and model_type is "TCN":
# Output weights at the first layer
plt.figure(2, figsize=(15, 15))
ws = model.get_weights()[0]
for i in range(min(36, len(ws.T))):
plt.subplot(6, 6, i + 1)
# imshow_(model.get_weights()[0][i][:,:,0]+model.get_weights()[1][i])
imshow_(np.squeeze(ws[:, :, :, i]).T)
res = sio.loadmat(result_path)
P_test = res['P'][0]
y_test = res['Y'][0]
max_classes = data.n_classes - 1
n_classes = data.n_classes
n_test = len(X_test)
print('- dataset: {}'.format(dataset))
print('- max_classes: {}'.format(max_classes))
# # Output all truth/prediction pairs
plt.figure(split, figsize=(20, 10))
P_test_ = np.array(P_test) / float(n_classes - 1)
y_test_ = np.array(y_test) / float(n_classes - 1)
for i in range(len(y_test)):
P_tmp = np.vstack([y_test_[i], P_test_[i]])
print(P_tmp.shape)
plt.subplot(n_test, 1, i + 1)
imshow_(P_tmp, vmin=0.0, vmax=1.0)
plt.xticks([])
plt.yticks([])
acc = np.mean(y_test[i] == P_test[i]) * 100
plt.ylabel("{:.01f}".format(acc))
# plt.title("Acc: {:.03}%".format(100*np.mean(P_test[i]==y_test[i]))
plt.show()