run.py

#!/usr/bin/env python
# coding: utf-8

# In[ ]:


from  utils import *
from  encrypt import *
from  model import *
from preprocess import *
from expansion import *
from generator import *
from const import *
import numpy as np
import random
import os
os.environ["CUDA_VISIBLE_DEVICES"] = "1" 
path_dataset =  'training_test_dataset.mat'


# In[ ]:



def train(model):
    for rounds in range(EPOCH):
        alluserembs=userembedding_layer.get_weights()[0]
        user_neighbor_emb=graph_embedding_expansion(Otraining,usernei,alluserembs)
        traingen=generate_batch_data_random(BATCH_SIZE,train_user_index,trainu,traini,usernei,trainlabel,user_neighbor_emb)
        cnt=0 
        batchloss=[]
        for i in traingen:
            layer_weights=model.get_weights()
            loss=model.train_on_batch(i[0],i[1])
            batchloss.append(loss)
            now_weights=model.get_weights() 
        
            sigma=np.std(now_weights[0]-layer_weights[0])
            #noise of pseudo interacted items
            norm=np.random.normal(0, sigma/np.sqrt(PSEUDO*BATCH_SIZE/now_weights[0].shape[0]), size=now_weights[0].shape) 
            now_weights[0]+=norm
            itemembedding_layer.set_weights([now_weights[0]])
            print(np.mean(batchloss))
            #ldp noise
            for i in range(len(now_weights)):
                now_weights[i]+=np.random.laplace(0,LR*2*CLIP/np.sqrt(BATCH_SIZE)/EPS,size=now_weights[i].shape)
            model.set_weights(now_weights)
            cnt+=1
            if cnt%10==0:
                print(cnt,loss)
            if cnt==len(train_user_index)//BATCH_SIZE:
                break
    return user_neighbor_emb

def test(model,user_neighbor_emb):            
    testgen=generate_batch_data(BATCH_SIZE,testu,testi,usernei,testlabel,user_neighbor_emb)
    cr = model.predict_generator(testgen, steps=len(testlabel)//BATCH_SIZE+1,verbose=1)
    print('rmse:',np.sqrt(np.mean(np.square(cr.flatten()-testlabel/LABEL_SCALE)))*LABEL_SCALE)
        
        


# In[ ]:


if __name__ == "__main__":
    
    #load data
    M = load_matlab_file(path_dataset, 'M')
    Otraining = load_matlab_file(path_dataset, 'Otraining')
    Otest = load_matlab_file(path_dataset, 'Otest')
    print('There are %i interactions logs.'%np.sum(np.array(np.array(M,dtype='bool'),dtype='int32')))
    
    #preprocess data    
    usernei=generate_history(Otraining)
    trainu,traini,trainlabel,train_user_index=generate_training_data(Otraining,M)
    testu,testi,testlabel=generate_test_data(Otest,M)
    
    #generate public&private keys     
    generate_key()
    
    #build model
    model,userembedding_layer,itemembedding_layer = get_model(Otraining)
    
    #train
    user_neighbor_emb = train(model)
    
    #test
    test(model,user_neighbor_emb)