不是phd的phd

# a = a + 4 * 5import torcha = torch.tensor([1,2,3])a.data.add_(torch.tensor(4),torch.tensor(5))print(a) # tensor([21, 22, 23])

# a = a + 4 / 2import torcha = torch.tensor([1,2,3])a.addcdiv(torch.tensor(4),torch.tensor(2))print(a) # tensor([1, 2, 3]) # 值不加a.data.addcdiv_(torch.tensor(4),torch.tensor(2))print(a) # tensor([3, 4, 5]) # 值加a = a.addcdiv(torch.tensor(4),torch.t

要先禁用nouveau保证lsmod |grep nouveau不显示信息

import torchtensor = torch.Tensor([[[1,1],[2,2],[3,3]],[[4,4],[5,5],[6,6]]])print(tensor.shape)print(tensor)pad_tensor = torch.constant_pad_nd(tensor,(0,0,0,2))print(pad_tensor.shape)print(pad_tensor)print结果：torch.Size([2, 3, 2])tensor([[[1., 1.]

label为[batch_size, num_class]logits为[batch_size, num_class]每个label为比如[0,0,1,0,0,0,1,0,1,0]，就是10类有3类正确不能用tf.nn.softmax_cross_entropy_with_logitsPytorch使用torch.nn.BCElossTensorflow使用tf.losses.sigmo...

https://github.com/pcyin/tranX/blob/master/model/pointer_net.py

https://github.com/pcyin/tranX/blob/master/model/seq2seq_copy.py

class Attention(nn.Module): """ inputs是[batch, seq_len1, hidden_dim] labels_num是seq_len2 """ def __init__(self, labels_num, hidden_size): super(Attention, self).__init__()...

import torchbatch_size = 2sequence_len = 3hidden_dim = 5x = torch.zeros(batch_size, sequence_len, hidden_dim).scatter_(dim=-1, index=torch.LongTensor([[[2],[2],[1]]...

很大可能是因为index的值超出了范围，比如import torchbatch_size = 2hidden_dim = 5x = torch.zeros(batch_size, hidden_dim).scatter_(dim=-1, index=torch.LongTensor([[2],[1]]), ...

import torchinput_tensor = torch.tensor([1,2,3,4,5])input_tensor = torch.tensor([[1,2,3,4,5],[6,7,8,9,10]])

import torch.nn.functional as F# p_logit: [batch,dim0]# q_logit: [batch,dim0]def kl_categorical(p_logit, q_logit): p = F.softmax(p_logit, dim=-1) _kl = torch.sum(p * (F.log_softmax(p_logit...

def kl_categorical(p_logit, q_logit): p = F.softmax(p_logit, dim=-1) _kl = torch.sum(p * (F.log_softmax(p_logit, dim=-1) - F.log_softmax(q_logit, dim=-1)), 1) retu...

def at_loss(embedder, encoder, clf, batch, perturb_norm_length=5.0): embedded = embedder(batch) # [seq_len,batch,hidden_dim] embedded.retain_grad() ce = F.cross_entropy((clf(encoder(embedd...

https://github.com/guotong1988/sqlova-debug-read/blob/master/sqlova/utils/utils_wikisql.py#L223

import torchinput_tensor = torch.tensor([[1,2],[3,4],[5,6]])gather_input = torch.tensor([[0,0],[1,0],[1,1]])output_tensor = torch.gather(input_tensor, 1, gather_input)print(output_tensor)tensor(...

import torchinput_tensor = torch.tensor([1,2,3,4,5])mask = input_tensor>3print(mask)indexes = mask.nonzero().squeeze()print(indexes)tensor([0, 0, 0, 1, 1], dtype=torch.uint8)tensor([3, 4])...

import torchinput_tensor = torch.tensor([1,2,3,4,5])print(input_tensor>3)mask = (input_tensor>3).nonzero()print(mask)print(input_tensor.index_select(0,mask))tensor([0, 0, 0, 1, 1], dtype=...

# p_logit: [batch,class_num]def entropy_loss(p_logit): p = F.softmax(p_logit, dim=-1) return -1 * torch.sum(p * F.log_softmax(p_logit, dim=-1)) / p_logit.size()[0]

/pytorch/aten/src/THC/THCTensorRandom.cuh:187: void sampleMultinomialOnce(long *, long, int, T *, T *, int, int) [with T = float, AccT = float]: block: [6,0,0], thread: [5,0,0] Assertion `THCNumerics&...

安装：其中requirements.txt里安装的是很新的torchtextgit clone --branch 0.9.1 https://github.com/OpenNMT/OpenNMT-py.gitcd OpenNMT-pypip install -r requirements.txt cd ..预处理：其中src-train.txt和tgt-train.txt为原始英文...

x = torch.Tensor(2,3)y = x.permute(1,0)y.view(-1) # 报错，因为x和y指针指向相同y = x.permute(1,0).contiguous()y.view(-1) # OK

import torchbatch_size = 2hidden_dim = 5x = torch.zeros(batch_size, hidden_dim).scatter_(dim=-1, index=torch.LongTensor([[2],[1]]), va...

class SelfAttention(nn.Module): """ scores each element of the sequence with a linear layer and uses the normalized scores to compute a context over the sequence. """ def __init__(sel...

输入BERT的方式 https://github.com/naver/sqlova/blob/master/sqlova/utils/utils_wikisql.py在generate_inputs方法[CLS] question_word_1,question_word_2,...question_word_n [SEP] header_1 [SEP] header_2 [SEP] ... ...

# seq: [batch,sel_len,hidden_dim]# cond: [batch,hidden_dim]# lens: [batch] def attend(seq, cond, lens): """ attend over the sequences `seq` using the condition `cond`. """ scores = ...

import torch.nn.functional as Fdistance = F.pairwise_distance(rep_a, rep_b, p=2)其中rep_a和rep_a为[batch_size,hidden_dim]

摘自https://github.com/UKPLab/sentence-transformers/blob/master/sentence_transformers/losses.py

其中rep_a为[batch_size,hidden_dim]rep_b为[batch_size,hidden_dim]labels为[batch_size]摘自https://github.com/UKPLab/sentence-transformers/blob/master/sentence_transformers/models/TransformerModel.py...

import torchinput_tensor = torch.tensor([1,2,3,4,5])print(input_tensor.index_select(0,torch.tensor([0,2,4])))input_tensor = torch.tensor([[1,2,3,4,5],[6,7,8,9,10]])print(input_tensor.index_select...

>>> torch.full((2, 3), 3.141592)tensor([[ 3.1416, 3.1416, 3.1416], [ 3.1416, 3.1416, 3.1416]])

import torchA_idx = torch.LongTensor([0, 2]) # the index vectorB = torch.LongTensor([[1, 2, 3], [4, 5, 6]])C = B.index_select(1, A_idx)# 1 3# 4 6

import torchimport torch.nn.functional as Ffrom torch.autograd import *a = Variable(torch.FloatTensor([[0,0,0,0,0,0,90,100]]))b=F.softmax(a,-1)print(b.multinomial()) # 7 或 6print(b.multinomial

就是抽样a = torch.FloatTensor([[0,0,0,0,0,0,0,100]])b=softmax(a)b.multinomial()Variable containing: 7 [torch.LongTensor of size 1x1]b.multinomial()Variable containing: 7 [torch.LongTen

git clone https://github.com/pytorch/vision.git git checkout 0.2.0 pip install pillow python setup.py install

http://pytorch.org/docs/0.3.0/distributions.htmlprobs = policy_network(state)m = Categorical(probs)action = m.sample() # 抽样一个actionnext_state, reward = env.step(action) # 得到一个rewardloss = -m.l

permute>>> img_nhwc = torch.randn(10, 480, 640, 3)>>> img_nhwc.size()torch.Size([10, 480, 640, 3])>>> img_nchw = img_nhwc.permute(0, 3, 1, 2)>>> img_nchw.size()torch.Size([10, 3, 480, 640])

class Net(nn.Module): … model = Net() …model.train() # 把module设成训练模式，对Dropout和BatchNorm有影响model.eval() # 把module设置为预测模式，对Dropout和BatchNorm模块有影响

不是concat的意思import torcha = torch.ones([1,2])b=torch.ones([1,2])torch.stack([a,b],1)(0 ,.,.) = 1 1 1 1[torch.FloatTensor of size 1x2x2]

import torcha = torch.ones([1,2])b = torch.ones([1,2])torch.cat([a,b],1) 1 1 1 1[torch.FloatTensor of size 1x4]