import numpy as np np.array([1,1]) and np.array([[1,1]]) x = x.reshape((-1,)) # 拉平
from numpy import * v = array([1,2,3,4]) type() v.shape v.dtype v.ndim x = arange(0, 10, 1) #1-10 linspace(0, 10, 25) #[0,10] 25 x, y = mgrid[0:5, 0:5] #网格
random.rand(5,5) random.randn(5,5) zeros((3,3)) ones((3,3))
M = array([[1, 2], [3, 4]]) M[1,:] = 0 M[:,2] = -1
v*v, v+2 dot(v, v.T) Other tips is on the Mr.Bu's notebook
如果弄错了int和float，可能是错的
[1,2]的shape值(2,)，意思是一维数组，数组中有2个元素。
[[1],[2]]的shape值是(2,1)，意思是一个二维数组，每行有1个元素。
[[1,2]]的shape值是（1，2），意思是一个二维数组，每行有2个元素。 ！！！！矩阵相乘一定要用下面两种形式，要不然就会出问题
所以在初始化的时候也要注意： nn.b2 = np.zeros(nn.n_output_dim) nn.b2 = np.zeros((1, nn.n_output_dim)) 上面第一个是一维的，第二个是二维的
import torch as t
函数 功能 Tensor(*sizes) 基础构造函数 ones(*sizes) 全1Tensor zeros(*sizes) 全0Tensor eye(*sizes) 对角线为1，其他为0 arange(s,e,step 从s到e，步长为step linspace(s,e,steps) 从s到e，均匀切分成steps份 rand/randn(*sizes) 均匀/标准分布 normal(mean,std)/uniform(from,to) 正态分布/均匀分布 randperm(m) 随机排列
t.ones_like(n)
b.tolist() c.shape a = t.arange(0, 6) a.view(2, 3) #通过tensor.view方法可以调整tensor的形状 a[:2, 0:2] a > 1 a[a>1]
t.Tensor(numpy) x.squeeze().numpy() x.squeeze(1) x.unsqueeze(1) .view_as(x_mean)
x.mm(w) torch.matmul(x, w))
from torch.autograd import Variable x_train = np.array([[3.3], [4.4], [5.5], [6.71], [6.93], [4.168], [9.779], [6.182], [7.59], [2.167], [7.042], [10.791], [5.313], [7.997], [3.1]], dtype=np.float32) x_train = torch.from_numpy(x_train) x_train = Variable(x_train) #numpy to tensor to variable
x_train.data.numpy() #variable to numpy
x = Variable(torch.FloatTensor([2, 3]), requires_grad=True) k.backward(torch.FloatTensor([1, 0]), retain_graph=True)
∂n∂m0=w0∂n0∂m0+w1∂n1∂m0=2m0+0=2×2=4 ∂n∂m1=w0∂n0∂m1+w1∂n1∂m1=0+3m21=3×32=27 read the notebook of Mr.Bu(autograd) It's different when caculating the one numbers and the list of numpy
y_ = linear_model(x_train) loss = get_loss(y_, y_train)
w.grad.zero_() # 记得归零梯度 b.grad.zero_() # 记得归零梯度 loss.backward()
w.data = w.data - 1e-2 * w.grad.data # 更新 w b.data = b.data - 1e-2 * b.grad.data # 更新 b
(a > 0.5) * 1
net = nn.Sequential(
nn.Linear(64, 400),
nn.ReLU(),
nn.Linear(400, 200),
nn.ReLU(),
nn.Linear(200, 100),
nn.ReLU(),
nn.Linear(100, 50),
nn.ReLU(),
nn.Linear(50, 10)
)
out = net(n, m) #输入数据类型 (number of samples, number of neuron) 只有一个m， 图片需要展开
net = nn.Sequential(
nn.Conv2d(in_channels, out_channels, kernel_size=3, padding=1)
nn.ReLU(),
nn.Conv2d(out_channels, out_channels, kernel_size=3, padding=1)
nn.ReLU(),
nn.MaxPool2d(2, 2)
)
out = net(n, in_channels, x, y) #(numbers of samples, in_channesls, x of img, y of img) 需要输入图片的通道数和长宽
RNN 的输入形式，(length of sequence, batch, numbers of feature)