torch-inspect 0.0.3

简单示例

import torch.nn as nn
import torch.nn.functional as F
import torch_inspect as ti

class SimpleNet(nn.Module):
    def __init__(self):
        super(SimpleNet, self).__init__()
        self.conv1 = nn.Conv2d(1, 6, 3)
        self.conv2 = nn.Conv2d(6, 16, 3)
        self.fc1 = nn.Linear(16 * 6 * 6, 120)
        self.fc2 = nn.Linear(120, 84)
        self.fc3 = nn.Linear(84, 10)

    def forward(self, x):
        x = F.max_pool2d(F.relu(self.conv1(x)), (2, 2))
        x = F.max_pool2d(F.relu(self.conv2(x)), 2)
        x = x.view(-1, self.num_flat_features(x))
        x = F.relu(self.fc1(x))
        x = F.relu(self.fc2(x))
        x = self.fc3(x)
        return x

    def num_flat_features(self, x):
        size = x.size()[1:]
        num_features = 1
        for s in size:
            num_features *= s
        return num_features


  net = SimpleNet()
  ti.summary(net, (1, 32, 32))

将产生以下输出

----------------------------------------------------------------
        Layer (type)               Output Shape         Param #
================================================================
            Conv2d-1           [100, 6, 30, 30]              60
            Conv2d-2          [100, 16, 13, 13]             880
            Linear-3                 [100, 120]          69,240
            Linear-4                  [100, 84]          10,164
            Linear-5                  [100, 10]             850
================================================================
Total params: 81,194
Trainable params: 81,194
Non-trainable params: 0
----------------------------------------------------------------
Input size (MB): 0.39
Forward/backward pass size (MB): 6.35
Params size (MB): 0.31
Estimated Total Size (MB): 7.05
----------------------------------------------------------------

要程序化访问网络信息，存在 inspect 函数

info = ti.inspect(net, (1, 32, 32))
print(info)

[LayerInfo(name='Conv2d-1', input_shape=[100, 1, 32, 32], output_shape=[100, 6, 30, 30], trainable_params=60, non_trainable_params=0),
 LayerInfo(name='Conv2d-2', input_shape=[100, 6, 15, 15], output_shape=[100, 16, 13, 13], trainable_params=880, non_trainable_params=0),
 LayerInfo(name='Linear-3', input_shape=[100, 576], output_shape=[100, 120], trainable_params=69240, non_trainable_params=0),
 LayerInfo(name='Linear-4', input_shape=[100, 120], output_shape=[100, 84], trainable_params=10164, non_trainable_params=0),
 LayerInfo(name='Linear-5', input_shape=[100, 84], output_shape=[100, 10], trainable_params=850, non_trainable_params=0)]

安装

安装过程简单，只需

$ pip install torch-inspect

要求

• Python 3.6+

• PyTorch 1.0+

参考和感谢

此包基于 pytorch-summary 和 PyTorch 问题 。与 pytorch-summary 相比，pytorch-inspect 支持RNN/LSTMs，并提供对网络摘要信息的程序访问。由于其更模块化的结构和测试的存在，它更容易扩展并支持更多功能。

变更

0.0.3 (2019-09-22)

• 添加LSTM支持

• 修复了多输入/输出支持

• 添加了更多的网络测试用例

• 批量大小不再默认为-1

0.0.2 (2019-09-22)

• 增加了批归一化支持

• 移除了设备参数

0.0.1 (2019-09-1)

• 初始版本。