import random as rd import collections from types import new_class import numpy as np import scipy.sparse as sp from scipy.sparse import csr_matrix from parse import parse_args import time import torch from copy import deepcopy from tqdm import tqdm import matplotlib.pyplot as plt import pandas as pd from torch.utils.data import Dataset, DataLoader from reckit import randint_choice import os import bisect # Helper function used when loading data from files def helper_load(filename): user_dict_list = {} item_dict = set() with open(filename) as f: for line in f.readlines(): line = line.strip('\n').split(' ') if len(line) == 0: continue line = [int(i) for i in line] user = line[0] items = line[1:] item_dict.update(items) if len(items) == 0: continue user_dict_list[user] = items return user_dict_list, item_dict, def helper_load_train(filename): user_dict_list = {} item_dict = set() item_dict_list = {} trainUser, trainItem = [], [] with open(filename) as f: for line in f.readlines(): line = line.strip('\n').split(' ') # print(line) if len(line) == 0: continue line = [int(i) for i in line] user = line[0] items = line[1:] item_dict.update(items) # LightGCN trainUser.extend([user] * len(items)) trainItem.extend(items) if len(items) == 0: continue user_dict_list[user] = items for item in items: if item in item_dict_list.keys(): item_dict_list[item].append(user) else: item_dict_list[item] = [user] return user_dict_list, item_dict, item_dict_list, trainUser, trainItem # It loads the data and creates a train_loader class AbstractData: def __init__(self, args): self.path = args.data_path + args.dataset + '/cf_data/' self.train_file = self.path + 'train.txt' self.valid_file = self.path + 'valid.txt' self.test_file = self.path + 'test.txt' self.mix = True if 'mix' in args.dataset else False if(args.nodrop): self.train_nodrop_file = self.path + 'train_nodrop.txt' self.nodrop = args.nodrop self.candidate = args.candidate if(args.candidate): self.test_neg_file = self.path + 'test_neg.txt' self.batch_size = args.batch_size self.neg_sample = args.neg_sample self.device = torch.device(args.cuda) self.model_name = args.model_name self.user_pop_max = 0 self.item_pop_max = 0 self.infonce = args.infonce self.num_workers = args.num_workers self.dataset = args.dataset self.candidate = args.candidate # Number of total users and items self.n_users, self.n_items, self.n_observations = 0, 0, 0 self.users = [] self.items = [] self.population_list = [] self.weights = [] # List of dictionaries of users and its observed items in corresponding dataset # {user1: [item1, item2, item3...], user2: [item1, item3, item4],...} # {item1: [user1, user2], item2: [user1, user3], ...} self.train_user_list = collections.defaultdict(list) self.valid_user_list = collections.defaultdict(list) if(self.dataset == "tencent_synthetic" or self.dataset == "kuairec_ood"): self.test_ood_user_list_1 = collections.defaultdict(list) self.test_ood_user_list_2 = collections.defaultdict(list) self.test_ood_user_list_3 = collections.defaultdict(list) else: self.test_user_list = collections.defaultdict(list) # Used to track early stopping point self.best_valid_recall = -np.inf self.best_valid_epoch, self.patience = 0, 0 self.train_item_list = collections.defaultdict(list) self.Graph = None self.trainUser, self.trainItem, self.UserItemNet = [], [], [] self.n_interactions = 0 if(self.dataset == "tencent_synthetic" or self.dataset == "kuairec_ood"): self.test_ood_item_list_1 = [] self.test_ood_item_list_2 = [] self.test_ood_item_list_3 = [] else: self.test_item_list = [] #Dataloader self.train_data = None self.train_loader = None self.load_data() # model-specific attributes self.add_special_model_attr(args) self.get_dataloader() def add_special_model_attr(self, args): pass # self.trainUser and self.trainItem are respectively the users and items in the training set, in the form of an interaction list. def load_data(self): self.train_user_list, train_item, self.train_item_list, self.trainUser, self.trainItem = helper_load_train( self.train_file) self.valid_user_list, valid_item = helper_load(self.valid_file) self.test_user_list, self.test_item_list = helper_load(self.test_file) if(self.nodrop): self.train_nodrop_user_list, self.train_nodrop_item_list = helper_load(self.train_nodrop_file) if(self.candidate): self.test_neg_user_list, self.test_neg_item_list = helper_load(self.test_neg_file) else: self.test_neg_user_list, self.test_neg_item_list = None, None self.pop_dict_list = [] temp_lst = [train_item, valid_item, self.test_item_list] self.users = list(set(self.train_user_list.keys())) self.items = list(set().union(*temp_lst)) self.items.sort() # print(self.items) self.n_users = len(self.users) self.n_items = len(self.items) print("n_users: ", self.n_users) print("n_items: ", self.n_items) for i in range(self.n_users): self.n_observations += len(self.train_user_list[i]) self.n_interactions += len(self.train_user_list[i]) if i in self.valid_user_list.keys(): self.n_interactions += len(self.valid_user_list[i]) if(self.dataset == "tencent_synthetic" or self.dataset == "kuairec_ood"): if i in self.test_ood_user_list_1.keys(): self.n_interactions += len(self.test_ood_user_list_1[i]) if i in self.test_ood_user_list_2.keys(): self.n_interactions += len(self.test_ood_user_list_2[i]) if i in self.test_ood_user_list_3.keys(): self.n_interactions += len(self.test_ood_user_list_3[i]) else: if i in self.test_user_list.keys(): self.n_interactions += len(self.test_user_list[i]) # Population matrix pop_dict = {} for item, users in self.train_item_list.items(): pop_dict[item] = len(users) + 1 for item in range(0, self.n_items): if item not in pop_dict.keys(): pop_dict[item] = 1 self.population_list.append(pop_dict[item]) pop_user = {key: len(value) for key, value in self.train_user_list.items()} pop_item = {key: len(value) for key, value in self.train_item_list.items()} self.pop_item = pop_item self.pop_user = pop_user # Convert to a unique value. sorted_pop_user = list(set(list(pop_user.values()))) sorted_pop_item = list(set(list(pop_item.values()))) sorted_pop_user.sort() sorted_pop_item.sort() self.n_user_pop = len(sorted_pop_user) self.n_item_pop = len(sorted_pop_item) user_idx = {} item_idx = {} for i, item in enumerate(sorted_pop_user): user_idx[item] = i for i, item in enumerate(sorted_pop_item): item_idx[item] = i self.user_pop_idx = np.zeros(self.n_users, dtype=int) self.item_pop_idx = np.zeros(self.n_items, dtype=int) # Convert the originally sparse popularity into dense popularity. for key, value in pop_user.items(): self.user_pop_idx[key] = user_idx[value] for key, value in pop_item.items(): # print(key, value) self.item_pop_idx[key] = item_idx[value] user_pop_max = max(self.user_pop_idx) item_pop_max = max(self.item_pop_idx) self.user_pop_max = user_pop_max self.item_pop_max = item_pop_max self.sample_items = np.array(self.items, dtype=int) if(self.mix): self.add_mixed_data() else: self.selected_train, self.selected_valid, self.selected_test = [], [], [] self.nu_info = [] self.ni_info = [] def add_mixed_data(self): self.selected_train, self.selected_valid, self.selected_test = [], [], [] self.nu_info = [] self.ni_info = [] self.mixed_datasets = ['movie', 'book', 'game'] # self.mixed_datasets = ['movie', 'book', 'game', 'electronic'] for data_name in self.mixed_datasets: train_train_file_, valid_file_, test_file_ = self.path + 'train_' + data_name + '.txt', self.path + 'valid_' + data_name + '.txt', self.path + 'test_' + data_name + '.txt' train_user_list_, train_item_, __, ___, ____ = helper_load_train(train_train_file_) valid_user_list_, valid_item_ = helper_load(valid_file_) test_user_list_, test_item_ = helper_load(test_file_) temp_lst = [train_item_, valid_item_, test_item_] users_ = list(set(train_user_list_.keys())) items_ = list(set().union(*temp_lst)) items_.sort() n_users_ = len(users_) n_items_ = len(items_) print(f"n_users_: {data_name}", n_users_) print(f"n_items_: {data_name}", n_items_) self.selected_train.append(train_user_list_) self.selected_valid.append(valid_user_list_) self.selected_test.append(test_user_list_) self.nu_info.append(n_users_) self.ni_info.append(n_items_) self.cum_ni_info = np.cumsum(self.ni_info) self.cum_ni_info = np.insert(self.cum_ni_info, 0, 0) self.cum_nu_info = np.cumsum(self.nu_info) self.cum_nu_info = np.insert(self.cum_nu_info, 0, 0) # self.train_file_movie, self.valid_file_movie, self.test_file_movie = self.path + 'train_movie.txt', self.path + 'valid_movie.txt', self.path + 'test_movie.txt' # self.train_user_list_movie, train_item_movie, __, ___, ____ = helper_load_train(self.train_file_movie) # self.valid_user_list_movie, valid_item_movie = helper_load(self.valid_file_movie) # self.test_user_list_movie, test_item_movie = helper_load(self.test_file_movie) # temp_lst = [train_item_movie, valid_item_movie, test_item_movie] # self.users_movie = list(set(self.train_user_list_movie.keys())) # self.items_movie = list(set().union(*temp_lst)) # self.items_movie.sort() # self.n_users_movie = len(self.users_movie) # self.n_items_movie = len(self.items_movie) # print("n_users_movie: ", self.n_users_movie) # print("n_items_movie: ", self.n_items_movie) # self.train_file_book, self.valid_file_book, self.test_file_book = self.path + 'train_book.txt', self.path + 'valid_book.txt', self.path + 'test_book.txt' # self.train_user_list_book, train_item_book, __, ___, ____ = helper_load_train(self.train_file_book) # self.valid_user_list_book, valid_item_book = helper_load(self.valid_file_book) # self.test_user_list_book, test_item_book = helper_load(self.test_file_book) # temp_lst = [train_item_book, valid_item_book, test_item_book] # self.users_book = list(set(self.train_user_list_book.keys())) # self.items_book = list(set().union(*temp_lst)) # self.items_book.sort() # self.n_users_book = len(self.users_book) # self.n_items_book = len(self.items_book) # print("n_users_book: ", self.n_users_book) # print("n_items_book: ", self.n_items_book) # self.train_file_game, self.valid_file_game, self.test_file_game = self.path + 'train_game.txt', self.path + 'valid_game.txt', self.path + 'test_game.txt' # self.train_user_list_game, train_item_game, __, ___, ____ = helper_load_train(self.train_file_game) # self.valid_user_list_game, valid_item_game = helper_load(self.valid_file_game) # self.test_user_list_game, test_item_game = helper_load(self.test_file_game) # temp_lst = [train_item_game, valid_item_game, test_item_game] # self.users_game = list(set(self.train_user_list_game.keys())) # self.items_game = list(set().union(*temp_lst)) # self.items_game.sort() # self.n_users_game = len(self.users_game) # self.n_items_game = len(self.items_game) # print("n_users_game: ", self.n_users_game) # print("n_items_game: ", self.n_items_game) # self.exclude_items = [self.items_movie, self.items_book] # self.selected_train = [self.train_user_list_movie, self.train_user_list_book] # self.nui_info = [[self.n_users_movie, self.n_items_movie], [self.n_users_book, self.n_items_book]] # self.selected_train = [self.train_user_list_movie, self.train_user_list_book, self.train_user_list_game] # self.nui_info = [[self.n_users_movie, self.n_items_movie], [self.n_users_book, self.n_items_book], [self.n_users_game, self.n_items_game]] def get_dataloader(self): self.train_data = TrainDataset(self.model_name, self.users, self.train_user_list, self.user_pop_idx, self.item_pop_idx, \ self.neg_sample, self.n_observations, self.n_items, self.sample_items, self.infonce, self.items, self.nu_info, self.ni_info) self.train_loader = DataLoader(self.train_data, batch_size=self.batch_size, shuffle=True, num_workers=self.num_workers, drop_last=True) def _convert_sp_mat_to_sp_tensor(self, X): coo = X.tocoo().astype(np.float32) row = torch.Tensor(coo.row).long() col = torch.Tensor(coo.col).long() index = torch.stack([row, col]) data = torch.FloatTensor(coo.data) return torch.sparse.FloatTensor(index, data, torch.Size(coo.shape)) def getSparseGraph(self): if self.Graph is None: try: pre_adj_mat = sp.load_npz(self.path + '/s_pre_adj_mat.npz') print("finish loading adjacency matrix") norm_adj = pre_adj_mat # If there is no preprocessed adjacency matrix, generate one. except: print("generating adjacency matrix") s = time.time() adj_mat = sp.dok_matrix((self.n_users + self.n_items, self.n_users + self.n_items), dtype=np.float32) adj_mat = adj_mat.tolil() self.trainItem = np.array(self.trainItem) self.trainUser = np.array(self.trainUser) self.UserItemNet = csr_matrix((np.ones(len(self.trainUser)), (self.trainUser, self.trainItem)), shape=(self.n_users, self.n_items)) R = self.UserItemNet.tolil() adj_mat[:self.n_users, self.n_users:] = R adj_mat[self.n_users:, :self.n_users] = R.T adj_mat = adj_mat.tocsr() sp.save_npz(self.path + '/adj_mat.npz', adj_mat) print("successfully saved adj_mat...") adj_mat = adj_mat.todok() rowsum = np.array(adj_mat.sum(axis=1)) d_inv = np.power(rowsum, -0.5).flatten() d_inv[np.isinf(d_inv)] = 0. d_mat = sp.diags(d_inv) norm_adj = d_mat.dot(adj_mat) norm_adj = norm_adj.dot(d_mat) norm_adj = norm_adj.tocsr() end = time.time() print(f"costing {end - s}s, saved norm_mat...") sp.save_npz(self.path + '/s_pre_adj_mat.npz', norm_adj) self.Graph = self._convert_sp_mat_to_sp_tensor(norm_adj) self.Graph = self.Graph.coalesce().cuda(self.device) return self.Graph class TrainDataset(torch.utils.data.Dataset): def __init__(self, model_name, users, train_user_list, user_pop_idx, item_pop_idx, neg_sample, \ n_observations, n_items, sample_items, infonce, items, nu_info = None, ni_info = None): self.model_name = model_name self.users = users self.train_user_list = train_user_list self.user_pop_idx = user_pop_idx self.item_pop_idx = item_pop_idx self.neg_sample = neg_sample self.n_observations = n_observations self.n_items = n_items self.sample_items = sample_items self.infonce = infonce self.items = items self.nu_info = nu_info self.ni_info = ni_info self.cum_ni_info = np.cumsum(self.ni_info) self.cum_ni_info = np.insert(self.cum_ni_info, 0, 0) self.cum_nu_info = np.cumsum(self.nu_info) self.cum_nu_info = np.insert(self.cum_nu_info, 0, 0) def __getitem__(self, index): index = index % len(self.users) user = self.users[index] if self.train_user_list[user] == []: pos_items = 0 else: pos_item = rd.choice(self.train_user_list[user]) user_pop = self.user_pop_idx[user] pos_item_pop = self.item_pop_idx[pos_item] if self.infonce == 1 and self.neg_sample == -1: #in-batch return user, pos_item, user_pop, pos_item_pop elif self.infonce == 1 and self.neg_sample != -1: # InfoNCE negative sampling if(len(self.nu_info) > 0): # period = index period = bisect.bisect_right(self.cum_nu_info, index) - 1 # print(self.cum_ni_info) exclude_items = list(np.array(self.train_user_list[user]) - self.cum_ni_info[period]) # print('perirod', period) # print("********************") # print('info', self.ni_info[period]) # neg_items = [1] neg_items = randint_choice(self.ni_info[period], size=self.neg_sample, exclusion=exclude_items) neg_items = list(np.array(neg_items) + self.cum_ni_info[period]) # if(index < self.nui_info[0][0]): # neg_items = randint_choice(self.nui_info[0][1], size=self.neg_sample, exclusion=self.train_user_list[user]) # elif(index < self.nui_info[0][0] + self.nui_info[1][0]): # # neg_items = randint_choice(self.n_items, size=self.neg_sample, exclusion=self.train_user_list[user]+self.exclude_items[0]) # exclude_items = list(np.array(self.train_user_list[user]) - self.nui_info[0][1]) # neg_items = randint_choice(self.nui_info[1][1], size=self.neg_sample, exclusion=exclude_items) # neg_items = list(np.array(neg_items) + self.nui_info[0][1]) # else: # exclude_items = list(np.array(self.train_user_list[user]) - self.nui_info[0][1] - self.nui_info[1][1]) # neg_items = randint_choice(self.nui_info[2][1], size=self.neg_sample, exclusion=exclude_items) # neg_items = list(np.array(neg_items) + self.nui_info[0][1] + self.nui_info[1][1]) else: neg_items = randint_choice(self.n_items, size=self.neg_sample, exclusion=self.train_user_list[user]) neg_items_pop = self.item_pop_idx[neg_items] return user, pos_item, user_pop, pos_item_pop, torch.tensor(neg_items).long(), neg_items_pop else: # BPR negative sampling. (only sample one negative item) while True: idx = rd.randint(0, self.n_items -1) neg_item = self.items[idx] if neg_item not in self.train_user_list[user]: break neg_item_pop = self.item_pop_idx[neg_item] return user, pos_item, user_pop, pos_item_pop, neg_item, neg_item_pop def __len__(self): return self.n_observations