def predict():
if request.method == 'POST':
try:
data = request.get_json()
f = open('my_classifier.pickle', 'rb')
classifier = pickle.load(f)
f.close()
except ValueError:
return jsonify("Please enter a mail.")
return jsonify(classifier.classify(get_features(data["mail"])))
embedding_dim = 300
word_indexes = get_word_indexes(word_indexes_filepath)
model = create_model_3(max_caption_len, word_indexes, embedding_dim, feature_dim, embedding_matrix_filepath)
checkpointer = ModelCheckpoint(filepath="model_weights.hdf5", verbose=0)
# model.load_weights('model_weights.hdf5')
while True:
for i in range(1, 8):
captions = 'dataset/frequencies/train_captions_' + str(i) + '.npy'
features = 'dataset/frequencies/train_features_' + str(i) + '.npy'
print(captions)
print(features)
train_captions = get_captions(captions)
train_features = get_features(features)
X_captions = train_captions[:, :-1]
Y = np.expand_dims(train_captions[:, 1:], -1)
X = [train_features, X_captions]
try:
model.fit(X, Y, nb_epoch=1, callbacks=[checkpointer], shuffle=True)
except KeyboardInterrupt:
checkpointer.on_epoch_end(0)
val_features_filepath = 'dataset/original/merged_val.npy'
processed_train_captions_filepath = 'dataset/frequencies/train_captions_7.npy'
processed_train_features_filepath = 'dataset/frequencies/train_features_7.npy'
processed_val_captions_filepath = 'dataset/frequencies/val_captions.npy'
processed_val_features_filepath = 'dataset/frequencies/val_features.npy'
embeddings_filepath = 'embeddings/glove.6B.300d.txt'
embedding_matrix_filepath = 'layers/embedding_matrix.npy'
embedding_dim = 300
max_caption_len = 19
train_captions = get_captions(train_captions_filepath)
train_features = get_features(train_features_filepath)
train_captions = train_captions[120000:]
train_features = train_features[120000:]
# val_captions = get_captions(val_captions_filepath)
# val_features = get_features(val_features_filepath)
add_sentence_tokens(train_captions)
# add_sentence_tokens(val_captions)
# word_frequencies = calculate_word_frequencies(train_captions, val_captions)
# with open('dataset/frequencies/word_frequencies.json', 'w') as f:
# json.dump(word_frequencies, f)
with open('dataset/frequencies/word_frequencies.json', 'r') as f:
def main(args):
print "Loading model"
model = models_dict[args.model_name].build()
model.load_weights(args.model_weights)
vc = init_vc(args.vc_input)
pygame.init()
pos_corners = calibration(vc)
# Start the diff process
prev_black = set()
prev_white = set()
init = False
moves_list = []
game_over = False
while not game_over:
frame = np.swapaxes(read_vc(vc),0,1)
t0 = time.time()
features, new_image = get_features(frame, pos_corners, height=562)
t1 = time.time()
if not init:
pygame.init()
width, height = new_image.shape[:2]
screen = pygame.display.set_mode((args.screen_res,args.screen_res))
init = True
classes = model.predict(features).argmax(axis=1)
t2 = time.time()
black = set([(i%19,i/19) for i,cl in enumerate(classes) if cl==2])
white = set([(i%19,i/19) for i,cl in enumerate(classes) if cl==1])
# get diff
diff_error = False
try:
col,move = diff_move((prev_black,prev_white),(black,white))
except DiffError:
diff_error = True
col, move = None, None
screen.fill((0,0,0))
global_surf = pygame.Surface((2*width,2*width))
# Display camera
surface = pygame.pixelcopy.make_surface(np.swapaxes(new_image, 0, 1))
global_surf.blit(surface,(width,width))
# Display previous
surface = get_board_surface(540, prev_black, prev_white,clear=False)
global_surf.blit(surface,(width,0))
# Display computer vision
surface = get_board_surface(540, black, white,clear=False,color=(255,0,0) if diff_error else (0,255,0))
global_surf.blit(surface,(0,0))
pygame.transform.smoothscale(global_surf, (args.screen_res,args.screen_res), screen)
pygame.display.flip()
t3 = time.time()
print "Times: {:.2f}, {:.2f}, {:.2f}".format(t1-t0,t2-t1,t3-t2)
recalibrate = False
events = pygame.event.get()
for event in events:
if event.type == pygame.KEYDOWN and event.key==pygame.K_ESCAPE:
game_over = True
elif event.type == pygame.KEYDOWN and event.key==pygame.K_c:
recalibrate = True
elif event.type == pygame.KEYDOWN and not diff_error:
moves_list.append((col,move))
prev_black, prev_white = black, white
if recalibrate:
pygame.quit()
pygame.init()
pos_corners = calibration(vc)
init = False
sgf_string = gen_sgf_string(moves_list, args.name_black, args.name_white)
with open(args.folder+'/'+args.filename,'w') as f:
f.write(sgf_string)
processed_val_captions_filepath = 'dataset/processed/val_captions.npy'
processed_val_features_filepath = 'dataset/processed/val_features.npy'
embeddings_filepath = 'embeddings/glove.6B.300d.txt'
embedding_matrix_filepath = 'layers/embedding_matrix.npy'
vocab_size = 10000
embedding_dim = 300
max_caption_len = 19
# train_captions = get_captions(train_captions_filepath)
# train_features = get_features(train_features_filepath)
val_captions = get_captions(val_captions_filepath)
val_features = get_features(val_features_filepath)
# add_sentence_tokens(train_captions)
add_sentence_tokens(val_captions)
word_indexes = calculate_word_indexes(train_captions, val_captions)
with open('dataset/word_indexes.json', 'w') as f:
json.dump(word_indexes, f)
train_captions = None
# with open('dataset/word_indexes.json', 'r') as f:
# word_indexes = json.load(f)
# embedding_matrix = create_embedding_matrix(word_indexes, embedding_dim, embeddings_filepath)