ORGS_FUSION/LoadBestModel.py

import tensorflow as tf
from utils import *
from metrics import *
from models import GCN_Align
import gc
import time
import os


def loadBestModel():

    dir_best_model = os.getcwd()+"\\data1\\100000\\zh_en\\model.ckpt"
    sess = tf.Session()

    # Define placeholders
    num_supports = 1
    ph_ae = {
        'support': [tf.sparse_placeholder(tf.float32) for _ in range(num_supports)],
        'features': tf.sparse_placeholder(tf.float32), #tf.placeholder(tf.float32),
        'dropout': tf.placeholder_with_default(0., shape=()),
        'num_features_nonzero': tf.placeholder_with_default(0, shape=())
    }
    ph_se = {
        'support': [tf.sparse_placeholder(tf.float32) for _ in range(num_supports)],
        'features': tf.placeholder(tf.float32),
        'dropout': tf.placeholder_with_default(0., shape=()),
        'num_features_nonzero': tf.placeholder_with_default(0, shape=())
    }

    # some flags
    flags = tf.app.flags
    FLAGS = flags.FLAGS
    flags.DEFINE_string('lang', 'zh_en', 'Dataset string.')  # 'zh_en', 'ja_en', 'fr_en'
    flags.DEFINE_float('learning_rate', 0.01, 'Initial learning rate.')
    flags.DEFINE_integer('epochs', 500, 'Number of epochs to train.')
    flags.DEFINE_float('dropout', 0.3, 'Dropout rate (1 - keep probability).')
    flags.DEFINE_float('gamma', 3.0, 'Hyper-parameter for margin based loss.')
    flags.DEFINE_integer('k', 4, 'Number of negative samples for each positive seed.')
    flags.DEFINE_float('beta', 0.3, 'Weight for structure embeddings.')
    flags.DEFINE_integer('se_dim', 100, 'Dimension for SE.')
    flags.DEFINE_integer('ae_dim', 100, 'Dimension for AE.')
    flags.DEFINE_integer('seed', 9, 'Proportion of seeds, 3 means 30%')

    # data process
    adj, ae_input, train, test = load_data(FLAGS.lang)
    support = [preprocess_adj(adj)]

    # 把具体值赋给事先定义好的placeholder
    feed_dict_ae = construct_feed_dict(ae_input, support, ph_ae)
    feed_dict_ae.update({ph_ae['dropout']: FLAGS.dropout})
    feed_dict_se = construct_feed_dict(1.0, support, ph_se)
    feed_dict_se.update({ph_se['dropout']: FLAGS.dropout})

    # 负样本填充placeholder
    t = 0
    k = 0
    e = ae_input[2][0]
    L = np.ones((t, k))
    neg_left = L.reshape((t * k,))
    L = np.ones((t, k))
    neg2_right = L.reshape((t * k,))
    neg2_left = np.random.choice(e, t * k)
    neg_right = np.random.choice(e, t * k)
    feed_dict_ae.update({'neg_left:0': neg_left, 'neg_right:0': neg_right, 'neg2_left:0': neg2_left, 'neg2_right:0': neg2_right})
    feed_dict_se.update({'neg_left:0': neg_left, 'neg_right:0': neg_right, 'neg2_left:0': neg2_left, 'neg2_right:0': neg2_right})

    # Create model
    model_func = GCN_Align
    # attribute embedding model
    model_ae = model_func(ph_ae, input_dim=ae_input[2][1], output_dim=FLAGS.ae_dim, ILL=train, sparse_inputs=True, featureless=False, logging=True)
    # structure embedding model
    model_se = model_func(ph_se, input_dim=ae_input[2][0], output_dim=FLAGS.se_dim, ILL=train, sparse_inputs=False, featureless=True, logging=True)

    # load model
    saver = tf.train.Saver()
    saver.restore(sess, dir_best_model)

    # run the last layer, get vector
    # print(len(feed_dict_ae))
    # for i in feed_dict_ae.keys():
    #     print(i)
    vec_ae = sess.run(model_ae.outputs, feed_dict=feed_dict_ae)
    vec_se = sess.run(model_se.outputs, feed_dict=feed_dict_se)

    # 清内存
    print("清内存")
    del saver
    del model_ae
    del model_se
    del model_func
    del feed_dict_ae
    del feed_dict_se
    del adj
    del ae_input
    del train
    # del test
    del support
    del sess
    gc.collect()

    # print("AE")
    # get_hits(vec_ae, test)
    # print("SE")
    # get_hits(vec_se, test)
    # print("SE+AE")
    # get_combine_hits(vec_se, vec_ae, FLAGS.beta, test)
    #
    # calculate similarity
    # print("AE Similarity")
    # print(len(vec_ae), len(test))
    # predict(vec_ae, test)
    # print("SE Similarity")
    # predict(vec_se, test)
    # print("AE+SE Similarity")
    # predict(np.concatenate([vec_se*FLAGS.beta, vec_ae*(1.0-FLAGS.beta)], axis=1), test)

    print("Predict New Align Orgs")
    start_time = time.time()
    predict_new(np.concatenate([vec_se*FLAGS.beta, vec_ae*(1.0-FLAGS.beta)], axis=1))
    print("use time", time.time()-start_time)

    print("e"+str(FLAGS.epochs), "d"+str(FLAGS.dropout), "k"+str(FLAGS.k), "s"+str(FLAGS.seed),
          "lr"+str(FLAGS.learning_rate), "b"+str(FLAGS.beta))


if __name__ == '__main__':
    loadBestModel()