import torch import torch.optim as optim from torch.utils.data import DataLoader from tqdm import tqdm import numpy as np from word2vec.data_reader import DataReader, Word2vecDataset from word2vec.model import SkipGramModel class Word2VecTrainer: def __init__(self, input_file, output_file, emb_dimension=100, batch_size=32, window_size=5, iterations=5, initial_lr=0.001, min_count=12): self.data = DataReader(input_file, min_count) dataset = Word2vecDataset(self.data, window_size) self.dataloader = DataLoader(dataset, batch_size=batch_size, shuffle=False, num_workers=0, collate_fn=dataset.collate) self.output_file_name = output_file self.emb_size = len(self.data.word2id) self.emb_dimension = emb_dimension self.batch_size = batch_size self.iterations = iterations self.initial_lr = initial_lr self.skip_gram_model = SkipGramModel(self.emb_size, self.emb_dimension) self.use_cuda = torch.cuda.is_available() self.device = torch.device("cuda" if self.use_cuda else "cpu") if self.use_cuda: self.skip_gram_model.cuda() def train(self): for iteration in range(self.iterations): print("\n\n\nIteration: " + str(iteration + 1)) optimizer = optim.SparseAdam(self.skip_gram_model.parameters(), lr=self.initial_lr) scheduler = torch.optim.lr_scheduler.CosineAnnealingLR(optimizer, len(self.dataloader)) running_loss = 0.0 for i, sample_batched in enumerate(tqdm(self.dataloader)): #print('!!!!!!!!') #print(sample_batched) if len(sample_batched[0]) > 1: pos_u = sample_batched[0].to(self.device) pos_v = sample_batched[1].to(self.device) neg_v = sample_batched[2].to(self.device) #print('*****') print((pos_u)) #print(np.shape(pos_v)[0]) print((neg_v)) print(np.shape(pos_u)[0]) print(np.shape(pos_v)[0]) print(np.shape(neg_v)[0]) exit() scheduler.step() optimizer.zero_grad() loss,emb_mappings = self.skip_gram_model.forward(pos_u, pos_v, neg_v) loss.backward() optimizer.step() running_loss = running_loss * 0.9 + loss.item() * 0.1 if i > 0 and i % 500 == 0: print(" Loss: " + str(running_loss)) self.skip_gram_model.save_embedding(self.data.id2word, self.output_file_name) return emb_mappings def train_dp(self): for iteration in range(self.iterations): print("\n\n\nIteration: " + str(iteration + 1)) optimizer = optim.Adam(self.skip_gram_model.parameters(), lr=self.initial_lr) scheduler = torch.optim.lr_scheduler.CosineAnnealingLR(optimizer, len(self.dataloader)) print(len(self.dataloader)) C = 1 sigma = 10467#lastfm running_loss = 0.0 for i, sample_batched in enumerate(tqdm(self.dataloader)): #print('!!!!!!!!') if len(sample_batched[0]) > 1: pos_u = sample_batched[0].to(self.device) pos_v = sample_batched[1].to(self.device) neg_v = sample_batched[2].to(self.device) #print('*****') # print(pos_u) # print(pos_v) # print(neg_v) # print(np.shape(pos_u)) # print(np.shape(pos_v)) # print(np.shape(neg_v)) # exit() scheduler.step() optimizer.zero_grad() loss,emb_mappings = self.skip_gram_model.forward(pos_u, pos_v, neg_v) grads = [torch.zeros(p.shape).to(self.device) for p in self.skip_gram_model.parameters()] igrad = torch.autograd.grad(loss, self.skip_gram_model.parameters(), retain_graph=True) l2_norm = torch.tensor(0.0).to(self.device) for g in igrad: l2_norm += g.norm(2) ** 2 # l2_norm += g.sum().square().tolist() # print('time12:', int(time.time() / 1000)) l2_norm = l2_norm.sqrt() divisor = max(torch.tensor(1.0).to(self.device), l2_norm / C) for i in range(len(igrad)): grads[i] += igrad[i] / divisor for i in range(len(grads)): grads[i] += sigma * C * (torch.randn_like(grads[i]).to(self.device)) grads[i] /= np.shape(pos_u)[0]+np.shape(neg_v)[0] p_list = [p for p in self.skip_gram_model.parameters()] for i in range(len(p_list)): p_list[i].grad = grads[i] #p_list[i].grad.detach_() print(p_list) loss.backward() optimizer.step() running_loss = running_loss * 0.9 + loss.item() * 0.1 if i > 0 and i % 500 == 0: print(" Loss: " + str(running_loss)) self.skip_gram_model.save_embedding(self.data.id2word, self.output_file_name) return emb_mappings if __name__ == '__main__': w2v = Word2VecTrainer(input_file="input.txt", output_file="out.vec") w2v.train()