WEIGHT_DECAY = 0.01 D_LEARNING_RATE = 0.01 G_LEARNING_RATE = 0.01 BETA = OUTPUT_DIM / 8.0 GAMMA = 0.1 WORKDIR = '/cache/' DIS_MODEL_BEST_FILE = '/cache/flickr_dis_teacher_modaLoss_' + str( OUTPUT_DIM) + '.model' DIS_MODEL_PRETRAIN_FILE = '/cache/dis_baseline_pretrain_' + str( OUTPUT_DIM) + '.model' train_i2t, train_i2t_pos, train_i2t_neg, train_t2i, train_t2i_pos, train_t2i_neg = ut.load_all_query_url( ) feature_dict = ut.load_all_feature(WORKDIR) label_dict = ut.load_all_label(WORKDIR) def generate_samples(sess, generator, train_list, train_pos, train_neg, flag): data = [] for query in train_pos: pos_list = train_pos[query] candidate_neg_list = train_neg[query] candidate_list = train_list[query] random.shuffle(pos_list) random.shuffle(candidate_neg_list) random.shuffle(candidate_list) sample_size = int(len(candidate_list) / SAMPLERATIO) candidate_list = candidate_list[0:sample_size]
LAMBDA = 0 BETA = OUTPUT_DIM / 8.0 GAMMA = 0.1 # WORKDIR = '../mir/' DIS_MODEL_BEST_FILE = '/....../teacher_best_pretrain' + str( OUTPUT_DIM) + '.model' #DIS_MODEL_BEST_I2I_FILE = '/home/huhengtong/UKD/teacher_UGACH/OL_teacher_best_i2i_' + str(OUTPUT_DIM) + '.model' # DIS_MODEL_NEWEST_FILE = './model/dis_newest_nn_' + str(OUTPUT_DIM) + '.model' train_i2t, train_i2t_pos, train_i2t_neg, train_t2i, train_t2i_pos, train_t2i_neg = ut.load_all_query_url( ) # pdb.set_trace() feature_dict = ut.load_all_feature() label_dict = ut.load_all_label() #print(len(feature_dict), len(label_dict)) def generate_samples(train_pos, train_neg, flag): data = [] for query in train_pos: pos_list = train_pos[query] candidate_neg_list = train_neg[query] random.shuffle(pos_list) random.shuffle(candidate_neg_list) for i in range(SELECTNUM): data.append((query, pos_list[i], candidate_neg_list[i]))
GAMMA = 0.1 WORKDIR = '../mir/' GEN_MODEL_BEST_FILE = './model/gan_best_nn_' + str(OUTPUT_DIM) + '.model' DIS_MODEL_BEST_FILE = './model/dis_best_nn_' + str(OUTPUT_DIM) + '.model' GEN_MODEL_NEWEST_FILE = './model/gan_newest_nn_' + str(OUTPUT_DIM) + '.model' DIS_MODEL_NEWEST_FILE = './model/dis_newest_nn_' + str(OUTPUT_DIM) + '.model' DIS_MODEL_PRETRAIN_FILE = './model/dis_pretrain_nn_' + str( OUTPUT_DIM) + '.model' train_i2t, train_i2t_pos, train_i2t_neg, train_t2i, train_t2i_pos, train_t2i_neg, test_i2t, test_i2t_pos, test_t2i, test_t2i_pos = ut.load_all_query_url( WORKDIR + 'list/', CLASS_DIM) feature_dict = ut.load_all_feature(WORKDIR + 'list/', WORKDIR + 'feature/') label_dict = ut.load_all_label(WORKDIR + 'list/') record_file = open('record_' + str(OUTPUT_DIM) + '.txt', 'w') record_file.close() def generate_samples(sess, generator, train_list, train_pos, train_neg, flag): data = [] for query in train_pos: pos_list = train_pos[query] candidate_neg_list = train_neg[query] candidate_list = train_list[query] random.shuffle(pos_list) random.shuffle(candidate_neg_list)