def run(tweets_train, tweets_test, evaluation=False, retweets=True, k=3):
reload(statistics)
print('Computing features...')
stats = statistics.compute_statistics(tweets_train, tweets_test, k=k)
train_features, train_labels = statistics.prepare_train_features(tweets_train, stats, retweets)
test_features = statistics.prepare_test_features(tweets_test, stats, retweets)
if evaluation:
np.savetxt('eval_train_features.csv', train_features, delimiter=",")
np.savetxt('eval_train_labels.csv', train_labels, delimiter=",")
np.savetxt('eval_test_features.csv', test_features, delimiter=",")
else:
np.savetxt('train_features.csv', train_features, delimiter=",")
np.savetxt('train_labels.csv', train_labels, delimiter=",")
np.savetxt('test_features.csv', test_features, delimiter=",")
print('Learning model...')
lr_model = model.learn_model(train_features, train_labels)
probabilities = model.apply_model(test_features, lr_model)
print('Preparing solution...')
if evaluation:
solution.prepare_solutions_for_evaluation(tweets_test, probabilities[:, 0])
else:
solution.prepare_solutions(tweets_test, probabilities[:, 0])
def test(test_data, parameters):
NSS_scores = []
for fn, (image, fixations) in test_data.items():
prediction = apply_model(image, parameters)
NSS = calculate_NSS(prediction, fixations)
NSS_scores.append(NSS)
return np.mean(NSS_scores)
def magic(self):
# do magic
self.image.save('image.jpg')
img = cv2.imread('image.jpg', 0)
img = crop(img)
prid = apply_model(img)
#print("predicted value is : ", prid)
my_label = tk.Label(root, text=prid, font=("Arial Bold", 150))
my_label.config(bg="black")
my_label.place(x=50, y=480, anchor='sw')
def display_NSS_calculation(image, fixations, parameters, figure=None, title=None):
prediction = apply_model(image, parameters)
plt.figure(1, (15,6))
plt.subplot(1,3,1)
plt.title('(a) Stimulus')
plt.imshow(image)
# want hunk_size sized bins; want to cover full data range
normalized = (prediction-tf.reduce_mean(prediction)) / tf.math.reduce_std(prediction)
NSS_values = [normalized[x, y] for x, y in fixations]
NSS = np.mean(NSS_values)
hunk_size = 1
smallest = np.min(normalized)//hunk_size * hunk_size
biggest = (np.max(normalized)//hunk_size + 2) * hunk_size
hunks = np.arange(smallest,biggest,hunk_size)
axes = plt.subplot(1,3,2)
plt.title('(b) Predicted salience map (normalized)\n and actual fixation points')
plt.imshow(normalized, cmap='Greys_r', vmin=-np.max(np.abs(normalized)), vmax=np.max(np.abs(normalized)))
contour = plt.contour(np.flip(normalized, axis=0), hunks, colors='k', linewidths=1, origin='image')
if len(fixations):
plt.scatter(fixations[:,1], fixations[:,0], marker='.')
axes.clabel(contour, fmt='%1d', colors='#006600', inline=1, fontsize=9)
axes = plt.subplot(1,3,3)
plt.title(f'(c) Calculating normalized scanpath saliency:\nhistogram and mean (NSS = {NSS:.2f})')
plt.hist(NSS_values, bins=hunks)
axes.axvline(x=NSS, color='black', linestyle='--')
axes.set_xticks(hunks)
axes.set_xticklabels([str(i) for i in hunks])
axes.set_aspect(1.0/axes.get_data_ratio()) # unfortunately needed for square subplot
if title is not None:
plt.suptitle(f'{title}\n(NSS: {NSS:.4f})')
if figure is not None:
plt.savefig(figure, bbox_inches='tight')
else:
plt.show()
plt.close()
return NSS
def __call__(self, image, parameters):
return apply_model(image, parameters, js_safe=True)
def parameter_gradient(image_sample, fixations, parameters):
prediction = apply_model(image_sample, parameters)
NSS = calculate_NSS(prediction, fixations)
# gradients will return a list with a single element: [d(-NSS)/d(parameters)]
grad = tf.gradients(-NSS, [parameters])[0]
return NSS, grad
np.save('params.npy', parameters.numpy())
else:
parameters.assign(np.load('params.npy'))
if testing:
print('Test NSS:', test(test_data, parameters))
NSS_scores = []
if len(sys.argv) > 1:
for filename in sys.argv[1:]:
unresized_image = tf.image.decode_image(
open(filename, 'rb').read())
image = tf.image.resize(unresized_image, (224, 224),
antialias=True) / 255
prediction = apply_model(image, parameters)
safe_filename = filename.replace('/', '_')
np.save('out/' + safe_filename + '.npy', prediction.numpy())
display_NSS_calculation(image, [],
parameters,
figure='out/' + safe_filename + '.pdf',
title=filename)
else:
all_NSS = []
for filename in test_data.keys():
image, fixations = test_data[filename]
figure_out = 'out/' + filename + '.pdf'
NSS = display_NSS_calculation(image,
fixations,
parameters,
import os
import json
from model import apply_model, load_model
models = os.listdir('../models')
input_string = input()
for i in models:
print(i)
model = load_model(os.path.join('../models', i))
data = json.loads(input_string)
pred, time = apply_model(model, data)
print(time)