- 博客(199)
- 资源 (12)
- 问答 (1)
- 收藏
- 关注
RSS订阅原创 toxic comment classification 数据集
https://huggingface.co/transformers/installation.htmlhttps://leemeng.tw/attack_on_bert_transfer_learning_in_nlp.htmlhttps://zhuanlan.zhihu.com/p/49271699https://zhuanlan.zhihu.com/p/51413773https://tech.meituan.com/2019/11/14/nlp-bert-practice.htmlhtt
2020-08-27 20:20:52
1129
原创 pytorch torchtext.data.Field
https://blog.csdn.net/zwqjoy/article/details/86490098
2020-08-26 18:35:37
2546
原创 pytorch FC_classification
import torchimport torch.nn.functional as Fimport matplotlib.pyplot as plt# torch.manual_seed(1) # reproducible# make fake datan_data = torch.ones(100, 2)x0 = torch.normal(2*n_data, 1) # class0 x data (tensor), shape=(100, 2)y0 = torch.zer
2020-08-26 18:35:06
291
原创 pytorch FC_regression
import torchimport torch.nn.functional as Fimport matplotlib.pyplot as plt# torch.manual_seed(1) # reproducible# 生成数据x = torch.unsqueeze(torch.linspace(-1, 1, 100), dim=1) # x data (tensor), shape=(100, 1) ->(batch,dim)y = x.pow(2) + 0.2*torc
2020-08-26 18:34:46
144
原创 pytorch RNN_regression
import torchimport torch.nn as nnimport numpy as npimport matplotlib.pyplot as plt# show data# steps = np.linspace(0, np.pi*2, 100, dtype=np.float32) # 在[0,2*pi]区间内生成100个数,float32为转tensor方便# x_np = np.sin(steps) # 生成正弦函数# y_np = np.cos(steps) # 生成
2020-08-26 18:34:23
200
原创 pytorch orchvision.transforms.Normalize
APICLASS torchvision.transforms.Normalize(mean, std, inplace=False)input需要为(C,H,W)，一般会用ToTensor()预处理对每个channel，做正态化，input[channel] = (input[channel] - mean[channel]) / std[channel]参数描述mean (sequence)Sequence of means for each channel.std
2020-08-25 20:21:10
201
原创 pytorch torchvision.transforms.ToTensor
API将Shape=(HWC)的PIL.Image或者numpy.ndarray转换成shape=(CHW)的范围在[0.0,1.0]的torch.FloatTensor像素值范围从[0, 255]转成 [ .0 , 1.0 ]PIL Image 需要为 (L, LA, P, I, F, RGB, YCbCr, RGBA, CMYK, 1)这几种模式之一np.ndarray需要 dtype = np.uint8上述两点不满足时，不进行像素值范围归一化(no scaling)CLASS tor
2020-08-25 20:20:49
336
原创 pytorch torchvision.transforms.Resize
APICLASS torchvision.transforms.Resize(size, interpolation=2)参数描述size (sequence or int)如果size是sequence(h,w)，则输出为此尺寸：如果size是int，则暗比例缩放interpolation (int, optional)插值方法，默认是双线性参考：https://pytorch.org/docs/stable/torchvision/transforms.htm
2020-08-25 20:20:20
5050
原创 pytorch torchvision.transforms.CenterCrop
应用以图片中心进行裁剪import torchivision.transformsimport PIL.Image as Imageimport torchvision.transforms #读入图片image=Image.open("./test.png")crop_obj = torchvision.transforms.CenterCrop((224, 224)) #生成一个CenterCrop类的对象,用来将图片从中心裁剪成224*224image = crop_obj(i
2020-08-25 20:19:57
1997
原创 pytorch torchvision.transform.Compose
应用transforms.Compose([ transforms.CenterCrop(10), transforms.ToTensor(),])```# API将transforms组合起来```pythonCLASS torchvision.transforms.Compose(transforms)```参数|描述--|--transforms (list of Transform objects)|list of transforms to compose
2020-08-25 20:19:38
363
原创 TORCHVISION MODELS
随机weights创建modelimport torchvision.models as modelsresnet18 = models.resnet18()alexnet = models.alexnet()vgg16 = models.vgg16()squeezenet = models.squeezenet1_0()densenet = models.densenet161()inception = models.inception_v3()googlenet = models.goo
2020-08-24 20:15:21
337
原创 pytorch utils.model_zoo
应用APItorch.utils.model_zoo.load_url(url, model_dir=None, map_location=None, progress=True, check_hash=False, file_name=None)加载序列化的torch对象。如果对象已经存在model_dir，则直接加载。model_dir在<hub_dir>/checkpoints中，而hub_dir可以通过get_dir()获取。参数描述url (string
2020-08-24 20:13:49
1212
原创 pytorch utils.data.DataLoader
1.应用自动分batchAPICLASS torch.utils.data.DataLoader(dataset, batch_size=1, shuffle=False, sampler=None, batch_sampler=None, num_workers=0, collate_fn=None, pin_memory=False, drop_last=False, timeout=0, worker_init_fn=None, multiprocessing_context=None, gen
2020-08-24 20:13:36
173
原创 pytorch WHAT IS TORCH.NN REALLY?
手动创建神经网络import torchimport torch.nn as nnfrom pathlib import Pathimport requestsimport pickleimport gzipfrom matplotlib import pyplotimport numpy as npimport math# 下载数据DATA_PATH = Path("data")PATH = DATA_PATH / "mnist"PATH.mkdir(parents=True
2020-08-24 20:13:27
246
原创 A Comprehensive Introduction to Torchtext
1.应用x = torch.tensor([1, 2, 3, 4])torch.unsqueeze(x, 0)－－－－－－－－－－－－－-－－－－－－－－－－tensor([[ 1, 2, 3, 4]])torch.unsqueeze(x, 1)－－－－－－－－－－－－－－－－－－－－－－－tensor([[ 1], [ 2], [ 3], [ 4]])2.API在指定的位置添加新的维度，新老tensor数据相同参数描述
2020-08-24 20:13:12
228
原创 Latex 导数相关符号
应用1.求导描述样式公式微分dy\mathrm{d} ydy$\mathrm{d} y$一阶dydx\frac{\mathrm{d} y }{\mathrm{d} x}dxdy$\frac{\mathrm{d} y }{\mathrm{d} x}$n阶dnydxn\frac{\mathrm{d}^{n} y }{\mathrm{d} x^{n}}dxndny$\frac{\mathrm{d} y }{\mathrm{d} x}$2.偏导描
2020-08-24 20:12:54
4538
原创 Latex 矩阵
1.输入矩阵使用\begin array ccc和\end array参数输入矩阵$$\begin{array}{c} 1 & 0 & 0\\\\ 0 & 1 & 0\\\\ 0 & 0 & 1\\\\\end{array}$$100010001\begin{array}{c} 1 & 0 & 0\\\\ 0 & 1 & 0\\\\ 0 & 0 &
2020-08-24 20:12:45
1778
原创 LEARNING PYTORCH WITH EXAMPLES
以下面的神经网络为例：由损失函数(sse,平方差)是2*(y_pred-y)x与w乘积结果h,来反向求x梯度是：grad_h.dot(w.T)x与w成绩结果h,来反向求w梯度是：x.T.dot(grad_h)ReLU,反向题图是：grad_h_relu.copy()[h<0]=0https://www.geek-share.com/detail/2722319286.htmlhttps://www.cnblogs.com/alan-blog-TsingHua/p/9981522.html
2020-08-23 11:13:35
164
原创 pytorch optim.SGD
1.应用import torchimport torch.nn as nnoptimizer = torch.optim.SGD(model.parameters(), lr=0.1, momentum=0.9)optimizer.zero_grad()loss_fn(model(input), target).backward()optimizer.step()概念最简单的更新规则是Stochastic Gradient Descent (SGD)：weight = weight -
2020-08-23 11:12:40
1928
原创 pytorch 创建神经网络
标准的训练神经网络流程是：1.定义包含权重的神经网络2.遍历所有输入数据3.处理数据4.计算损失值5.向前传播6.更新权重参考：https://pytorch.org/tutorials/beginner/blitz/neural_networks_tutorial.html#sphx-glr-beginner-blitz-neural-networks-tutorial-py...
2020-08-23 11:12:30
137
原创 pytorch nn.MSELoss
1.应用import torchimport torch.nn as nnloss = nn.MSELoss()input = torch.tensor([1.0,1.0], requires_grad=True)target = torch.tensor([7.0,9.0])output = loss(input, target) # 50output.backward()2.概念API类mean squared error (squared L2 norm)CLASS tor
2020-08-23 11:11:58
453
原创 pytorch Tensor autograd functions
属性方法grad默认为None，当调用backward()方法后，会保存gradient梯度requires_gradTrue时，表示该tensor需要计算gradients，添加到graph中is_leafrequires_grad=False都是leaf tensor；requires_grad=True时，如果是用户创建的，既没有经过计算的为leaf tensor，其余的都不是backward()detach()返回新的tensor，从当前的gra...
2020-08-23 11:11:45
124
原创 pytorch nn.Module.zero_grad
设置model parameters的gradients 为 0APIzero_grad() → None参考：https://pytorch.org/docs/stable/generated/torch.nn.Module.html#torch.nn.Module.zero_grad
2020-08-23 11:11:32
494
原创 pytorch Tensor.backward
1.应用import torchimport torch.nn as nn# 1.全部为1x = torch.tensor([1.0,3.0], requires_grad = True) # tensor([1., 3.], requires_grad=True)y = x*x # tensor([1., 9.], grad_fn=<MulBackward0>)y.backward(torch.ones(2)) # 1*x[0] + 1*x[1]x.grad## 等价于x
2020-08-23 11:11:09
470
原创 pytorch nn.Module.parameters
返回模型的parameters的迭代对象。1.应用>>> for param in model.parameters():>>> print(type(param), param.size())<class 'torch.Tensor'> (20L,)<class 'torch.Tensor'> (20L, 1L, 5L, 5L)APIparameters(recurse: bool = True) → Iterator[t
2020-08-23 11:10:48
1232
原创 pytorch Tensor
torch.Tensor是torch.FloatTensor的简写形式：1.从list和numpy创建tensortorch.tensor([[1., -1.], [1., -1.]])torch.tensor(np.array([[1, 2, 3], [4, 5, 6]]))2.指定tensor数据类型和设备torch.zeros([2, 4], dtype=torch.int32)cuda0 = torch.device('cuda:0')torch.ones([2, 4], dtyp
2020-08-22 17:42:21
155
原创 torch nn.MaxPool2d
1.应用import torchimport torch.nn as nnm = nn.MaxPool2d(2)input = torch.randn(1, 1, 4, 4)output = m(input)API1.MaxPool2d类CLASS torch.nn.MaxPool2d(kernel_size: Union[T, Tuple[T, ...]], stride: Optional[Union[T, Tuple[T, ...]]] = None, padding: Union
2020-08-22 17:42:09
712
原创 pytorch nn.ReLU
Applies the rectified linear unit function element-wise:1.应用import torchimport torch.nn as nnm = nn.ReLU()input = torch.tensor([-1,1]) # tensor([ -1, 1])output = m(input) # tensor([0, 1])2.APIReLU类CLASS torch.nn.ReLU(inplace: bool = False)
2020-08-22 17:41:56
1328
原创 pytorch torchtext
连乘积1.应用import torchimport torch.nn as nna = torch.randn(1, 2, 3)torch.numel(a) # 1*2*3 = 6a = torch.zeros(4,4)torch.numel(a) # 4*4 = 16APItorch.numel(input) → int参数描述input (Tensor)the input tensor.参考：https://pytorch.org/docs/sta
2020-08-22 17:41:46
144
原创 Language modeling tutorial in torchtext
改变tensor的shape。返回的tensor必须包含相同的数据，相同的数量。import torchimport torch.nn as nnx = torch.randn(4, 4)x.size() # torch.Size([4, 4])y = x.view(16)y.size() # torch.Size([16])z = x.view(-1, 8)z.size() # torch.Size([2, 8]) -1表示此维度根据其他维度计算得来，2=16/8a = torch
2020-08-22 17:41:22
157
原创 pytorch nn.Linear
用于设置网络中的全连接层的，全连接层的输入与输出都是二维张量，一般形状为[batch_size, size]，而卷积层的输入输出是四维张量[batch, channel , height , width]import torchimport torch.nn as nnm = nn.Linear(10, 20) # (inputfeatures,outputfeatures)input = torch.randn(1, 10) # (batch,features)output = m(input)
2020-08-22 17:41:08
144
原创 pytorch nn.Conv2d
import torchimport torch.nn as nnx = torch.randn(1, 1, 32, 32) # batch, channel , height , widthprint(x.shape) # torch.Size([1, 1, 32, 32])conv = nn.Conv2d(1, 1, (3, 3)) # in_channel, out_channel ,kennel_size,strideprint(conv) # Conv2d(1, 1, kernel_
2020-08-22 17:40:57
259
原创 pytorch nn.Conv1d
一维卷积nn.Conv1d一般用于文本数据1.应用import torchimport torch.nn as nnx = torch.randn(1, 1, 32) # batch, channel , widthprint(x.shape) # torch.Size([1, 1, 32])conv = nn.Conv1d(1, 1, 3) # in_channel, out_channel ,kennel_sizeprint(conv) # Conv1d(1, 1, kernel_si
2020-08-22 17:40:42
197
1
原创 pytorch AUTOGRAD
torch.Tensor实例中设置torch.Tensor=True将会记录所有的操作记录。当完成一系列计算后可以使用.backward()方法自动计算梯度。而这个梯度将会累加在.grad属性上。.detach()方法可以停止追踪历史，未来的计算也不会被记录。with torch.no_grad():可以实现.detach()。这对我们验证模型有很大的好处，因为在验证模型的时候，我们不需要计算梯度。1.1.创建import torchx = torch.ones(2, 2, requires_g
2020-08-22 17:40:28
107
原创 WHAT IS PYTORCH
pytorch是：1）可以使用GPU的Numpy2）深度学习的框架1.TensorsTensors类似于Numpy的ndarrays，区别在于Tensor可以使用GPU。创建5*3的矩阵，未初始化from __future__ import print_functionimport torchx = torch.empty(5, 3)print(x)----------------------------------------------tensor([[ 5.0375e+28,
2020-08-22 17:36:32
118
原创 mysql set
1.setSET @name = 43;SET @total_tax = (SELECT SUM(tax) FROM taxable_transactions);
2020-08-20 23:49:56
246
2016最新中国行政区划分
2016-12-12
CSDN 如何搜索自己感兴趣的问题回答？
2017-05-14
TA创建的收藏夹 TA关注的收藏夹
TA关注的人