- 以一维卷积神经网络为例
1 直接使用 numpy 与 tensor 来构建数据集
1.1一维卷积神经网络结构
class Net(nn.Module):
def __init__(self):
super(Net, self).__init__()
self.conv1 = nn.Conv1d(in_channels=1, out_channels=10, kernel_size=3, stride=2)
self.max_pool1 = nn.MaxPool1d(kernel_size=3, stride=2)
self.conv2 = nn.Conv1d(10, 20, 3, 2)
self.max_pool2 = nn.MaxPool1d(3, 2)
self.conv3 = nn.Conv1d(20, 40, 3, 2)
self.liner1 = nn.Linear(40 * 14, 120)
self.liner2 = nn.Linear(120, 84)
self.liner3 = nn.Linear(84, 4)
def forward(self, x):
x = F.relu(self.conv1(x))
x = self.max_pool1(x)
x = F.relu(self.conv2(x))
x = self.max_pool2(x)
x = F.relu(self.conv3(x))
x = x.view(-1, 40 * 14)
x = F.relu(self.liner1(x))
x = F.relu(self.liner2(x))
x = self.liner3(x)
return x
1.2 数据的结构
x_train, x_test, y_train, y_test = train_test_split(dataSet, labels, test_size=0.3, random_state=42)
print( x_train.shape, y_train.shape, x_test.shape, y_test.shape)
out:
((51527, 500), (51527,), (22084, 500), (22084,))
1.3 数据的结构
1.3.1 将 numpy 数据集转化为 tensor
x_train = torch.from_numpy(x_train)
y_train = torch.from_numpy(y_train)
x_test = torch.from_numpy(x_test)
y_test = torch.from_numpy(y_test)
1.3.2 训练集数据类型转化为：tensor.float32
x_train = torch.tensor(x_train, dtype=torch.float32)
x_test = torch.tensor(x_test, dtype=torch.float32)
1.3.3 改变 x_train.shape , x_test.shape 的形状
送入训练的数据格式为：（1， 1， 500）
x_train = x_train.reshape(x_train.shape[0], 1, 1, x_train.shape[1])
x_test = x_test.reshape(x_test.shape[0], 1, x_train.shape[1])
print(x_train.shape, x_test.shape)
out:
(torch.Size([51527, 1, 1, 500]), torch.Size([22084, 1, 500]))
1.3.4 标签的数据格式 与 数据类型
- 在1.2 节中给出标签的形状：
x_train.shape = (51527,) , y_train.shape = (22084,)，，类型是：float， 但是训练器需要的是：（51527, 1）, (22084, 1) ,dtype=tensor.long
y_train = y_train.reshape(y_train.shape[0], 1)
y_test = y_test.reshape(y_test.shape[0], 1)
y_train = torch.tensor(y_train, dtype=torch.long)
y_test = torch.tensor(y_test, dtype=torch.long)
1.4 开始训练
# 定义损失函数 与 优化器
criterion = nn.CrossEntropyLoss()
optimizer = optim.SGD(net.parameters(), lr=0.001, momentum=0.9)
start = time.time()
for epoch in tqdm(range(10)):
running_loss = 0
for i, input_data in enumerate(x_train, 0):
# print(input_data.shape)
label = y_train[i]
optimizer.zero_grad()
outputs = net(input_data)
loss = criterion(outputs, label)
loss.backward()
optimizer.step()
running_loss += loss.item()
if i % 2000 == 1999:
print('[%d, %5d] loss: %0.3f' % (epoch + 1, i + 1, running_loss / 2000))
running_loss = 0.0
print('time = %2dm:%2ds' % ((time.time() - start)//60, (time.time()-start)%60))
2 通过继承 torch.utils.data.Dataset类构建一个数据集
class NuclearDataset(Dataset):
""" Nuclear Dataset."""
def __init__(self, data_file, root_dir=None, transform=None):
"""
Args:
data_file (string): Path to the data file with annotations.
root_dir (string): Directory with all the images.
transform (callable, optional): Optional transform to be applied
on a sample.
"""
# self.landmarks_frame = pd.read_csv(csv_file)
# 加载数据
self.landmarks_frame = np.load(data_file)
self.root_dir = root_dir
self.transform = transform
def __len__(self):
return len(self.landmarks_frame)
def __getitem__(self, idx):
if torch.is_tensor(idx):
idx = idx.tolist()
# img_name = os.path.join(self.root_dir,self.landmarks_frame.iloc[idx, 0])
# image = io.imread(img_name)
# landmarks = self.landmarks_frame.iloc[idx, 1:]
landmarks = self.landmarks_frame[:, -1]
landmarks = np.array([landmarks])
# landmarks = landmarks.astype('float').reshape(-1, 2)
landmarks = landmarks.astype('float').reshape(-1, 1)
sample = {'landmarks': landmarks}
if self.transform:
sample = self.transform(sample)
return sample
dataSet实例
标签
同样可以直接定义函数 划分训练集与训练集，比较方便，代码可移植性较好。这其中数据的形状是怎样的，可以按照代码运行的错误提示进行修改。