Skip to content

Commit de0f925

Browse files
AsakusaRinneAsakusaRinne
committed
Add Chinese version readme.
1 parent 069951e commit de0f925

File tree

2 files changed

+218
-1
lines changed

2 files changed

+218
-1
lines changed

‎README.md

+3-1
Original file line numberDiff line numberDiff line change
@@ -9,12 +9,14 @@
99
[![Badge](https://img.shields.io/badge/link-996.icu-red.svg)](https://996.icu/#/en_US)
1010
[![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/javiercp/BinderTF.NET/master?urlpath=lab)
1111

12+
English | [中文](docs/Readme-CN.md)
13+
1214
*master branch is based on tensorflow v2.x, v0.6x branch is based on tensorflow v2.6, v0.15-tensorflow1.15 is from tensorflow1.15.*
1315

1416

1517
![tensors_flowing](docs/assets/tensors_flowing.gif)
1618

17-
### Why TensorFlow in C# and F# ?
19+
### Why TensorFlow.NET ?
1820

1921
`SciSharp STACK`'s mission is to bring popular data science technology into the .NET world and to provide .NET developers with a powerful Machine Learning tool set without reinventing the wheel. Since the APIs are kept as similar as possible you can immediately adapt any existing TensorFlow code in C# or F# with a zero learning curve. Take a look at a comparison picture and see how comfortably a TensorFlow/Python script translates into a C# program with TensorFlow.NET.
2022

‎docs/README-CN.md

+215
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,215 @@
1+
![logo](assets/tf.net.logo.png)
2+
3+
**Tensorflow.NET**是AI框架[TensorFlow](https://www.tensorflow.org/)在.NET平台上的实现，支持C#和F#，可以用来搭建深度学习模型并进行训练和推理，并内置了Numpy API，可以用来进行其它科学计算。
4+
5+
Tensorflow.NET并非对于Python的简单封装，而是基于C API的pure C#实现，因此使用时无需额外的环境，可以很方便地用NuGet直接安装使用。并且dotnet团队提供的[ML.NET](https://github.com/dotnet/machinelearning)也依赖于Tensorflow.NET，支持调用Tensorflow.NET进行训练和推理，可以很方便地融入.NET生态。
6+
7+
与tensorflow相同，Tensorflow.NET也内置了Keras这一高级API，只要在安装Tensorflow.NET的同时安装Tensorflow.Keras就可以使用，Keras支持以模块化的方式调用模型，给模型的搭建提供了极大的便利。
8+
9+
[![Join the chat at https://gitter.im/publiclab/publiclab](https://badges.gitter.im/Join%20Chat.svg)](https://gitter.im/sci-sharp/community)
10+
[![Tensorflow.NET](https://ci.appveyor.com/api/projects/status/wx4td43v2d3f2xj6?svg=true)](https://ci.appveyor.com/project/Haiping-Chen/tensorflow-net)
11+
[![NuGet](https://img.shields.io/nuget/dt/TensorFlow.NET.svg)](https://www.nuget.org/packages/TensorFlow.NET)
12+
[![Documentation Status](https://readthedocs.org/projects/tensorflownet/badge/?version=latest)](https://tensorflownet.readthedocs.io/en/latest/?badge=latest)
13+
[![Badge](https://img.shields.io/badge/link-996.icu-red.svg)](https://996.icu/#/en_US)
14+
[![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/javiercp/BinderTF.NET/master?urlpath=lab)
15+
16+
中文 | [English](https://github.com/SciSharp/TensorFlow.NET#readme)
17+
18+
*当前主分支与Tensorflow2.10版本相对应，支持Eager Mode，同时也支持v1的静态图。*
19+
20+
21+
![tensors_flowing](assets/tensors_flowing.gif)
22+
23+
### Why TensorFlow.NET?
24+
25+
`SciSharp STACK`开源社区的目标是构建.NET平台下易用的科学计算库，而Tensorflow.NET就是其中最具代表性的仓库之一。在深度学习领域Python是主流，无论是初学者还是资深开发者，模型的搭建和训练都常常使用Python写就的AI框架，比如tensorflow。但在实际应用深度学习模型的时候，又可能希望用到.NET生态，亦或只是因为.NET是自己最熟悉的领域，这时候Tensorflow.NET就有显著的优点，因为它不仅可以和.NET生态很好地贴合，其API还使得开发者很容易将Python代码迁移过来。下面的对比就是很好的例子，Python代码和C#代码有着高度相似的API，这会使得迁移的时候无需做过多修改。
26+
27+
![pythn vs csharp](assets/syntax-comparision.png)
28+
29+
除了高度相似的API外，Tensorflow.NET与tensorflow也已经打通数据通道，tensorflow训练并保存的模型可以在Tensorflow.NET中直接读取并继续训练或推理，反之Tensorflow.NET保存的模型也可以在tensorflow中读取，这大大方便了模型的训练和部署。
30+
31+
与其它类似的库比如[TensorFlowSharp](https://www.nuget.org/packages/TensorFlowSharp/)相比，Tensorflow.NET的实现更加完全，提供了更多的高级API，使用起来更为方便，更新也更加迅速。
32+
33+
34+
### 文档
35+
36+
基本介绍与简单用例：[Tensorflow.NET Documents](https://scisharp.github.io/tensorflow-net-docs)
37+
38+
详细文档：[The Definitive Guide to Tensorflow.NET](https://tensorflownet.readthedocs.io/en/latest/FrontCover.html)
39+
40+
例程：[TensorFlow.NET Examples](https://github.com/SciSharp/TensorFlow.NET-Examples)
41+
42+
运行例程常见问题：[Tensorflow.NET FAQ](tensorflowlib/README.md)
43+
44+
### 安装与使用
45+
46+
安装可以在NuGet包管理器中搜索包名安装，也可以用下面命令行的方式。
47+
48+
安装分为两个部分，第一部分是Tensorflow.NET的主体：
49+
50+
```sh
51+
### 安装Tensorflow.NET
52+
PM> Install-Package TensorFlow.NET
53+
### 安装Tensorflow.Keras
54+
PM> Install-Package TensorFlow.Keras
55+
```
56+
57+
第二部分是计算支持部分，只需要根据自己的设备和系统选择下面之一即可：
58+
59+
```
60+
### CPU版本
61+
PM> Install-Package SciSharp.TensorFlow.Redist
62+
63+
### Windows下的GPU版本（需要安装CUDA和CUDNN）
64+
PM> Install-Package SciSharp.TensorFlow.Redist-Windows-GPU
65+
66+
### Linux下的GPU版本（需要安装CUDA和CUDNN）
67+
PM> Install-Package SciSharp.TensorFlow.Redist-Linux-GPU
68+
```
69+
70+
下面给出两个简单的例子，更多例子可以在[TensorFlow.NET Examples]中查看。
71+
72+
#### 简单例子（使用Eager Mode进行线性回归）
73+
74+
```csharp
75+
using static Tensorflow.Binding;
76+
using static Tensorflow.KerasApi;
77+
using Tensorflow;
78+
using Tensorflow.NumPy;
79+
80+
// Parameters
81+
var training_steps = 1000;
82+
var learning_rate = 0.01f;
83+
var display_step = 100;
84+
85+
// Sample data
86+
var X = np.array(3.3f, 4.4f, 5.5f, 6.71f, 6.93f, 4.168f, 9.779f, 6.182f, 7.59f, 2.167f,
87+
7.042f, 10.791f, 5.313f, 7.997f, 5.654f, 9.27f, 3.1f);
88+
var Y = np.array(1.7f, 2.76f, 2.09f, 3.19f, 1.694f, 1.573f, 3.366f, 2.596f, 2.53f, 1.221f,
89+
2.827f, 3.465f, 1.65f, 2.904f, 2.42f, 2.94f, 1.3f);
90+
var n_samples = X.shape[0];
91+
92+
// We can set a fixed init value in order to demo
93+
var W = tf.Variable(-0.06f, name: "weight");
94+
var b = tf.Variable(-0.73f, name: "bias");
95+
var optimizer = keras.optimizers.SGD(learning_rate);
96+
97+
// Run training for the given number of steps.
98+
foreach (var step in range(1, training_steps + 1))
99+
{
100+
// Run the optimization to update W and b values.
101+
// Wrap computation inside a GradientTape for automatic differentiation.
102+
using var g = tf.GradientTape();
103+
// Linear regression (Wx + b).
104+
var pred = W * X + b;
105+
// Mean square error.
106+
var loss = tf.reduce_sum(tf.pow(pred - Y, 2)) / (2 * n_samples);
107+
// should stop recording
108+
// Compute gradients.
109+
var gradients = g.gradient(loss, (W, b));
110+
111+
// Update W and b following gradients.
112+
optimizer.apply_gradients(zip(gradients, (W, b)));
113+
114+
if (step % display_step == 0)
115+
{
116+
pred = W * X + b;
117+
loss = tf.reduce_sum(tf.pow(pred - Y, 2)) / (2 * n_samples);
118+
print($"step: {step}, loss: {loss.numpy()}, W: {W.numpy()}, b: {b.numpy()}");
119+
}
120+
}
121+
```
122+
123+
这一用例也可以在[Jupyter Notebook Example](https://github.com/SciSharp/SciSharpCube)进行运行.
124+
125+
#### 简单例子（使用Keras搭建Resnet）
126+
127+
```csharp
128+
using static Tensorflow.Binding;
129+
using static Tensorflow.KerasApi;
130+
using Tensorflow;
131+
using Tensorflow.NumPy;
132+
133+
var layers = new LayersApi();
134+
// input layer
135+
var inputs = keras.Input(shape: (32, 32, 3), name: "img");
136+
// convolutional layer
137+
var x = layers.Conv2D(32, 3, activation: "relu").Apply(inputs);
138+
x = layers.Conv2D(64, 3, activation: "relu").Apply(x);
139+
var block_1_output = layers.MaxPooling2D(3).Apply(x);
140+
x = layers.Conv2D(64, 3, activation: "relu", padding: "same").Apply(block_1_output);
141+
x = layers.Conv2D(64, 3, activation: "relu", padding: "same").Apply(x);
142+
var block_2_output = layers.Add().Apply(new Tensors(x, block_1_output));
143+
x = layers.Conv2D(64, 3, activation: "relu", padding: "same").Apply(block_2_output);
144+
x = layers.Conv2D(64, 3, activation: "relu", padding: "same").Apply(x);
145+
var block_3_output = layers.Add().Apply(new Tensors(x, block_2_output));
146+
x = layers.Conv2D(64, 3, activation: "relu").Apply(block_3_output);
147+
x = layers.GlobalAveragePooling2D().Apply(x);
148+
x = layers.Dense(256, activation: "relu").Apply(x);
149+
x = layers.Dropout(0.5f).Apply(x);
150+
// output layer
151+
var outputs = layers.Dense(10).Apply(x);
152+
// build keras model
153+
var model = keras.Model(inputs, outputs, name: "toy_resnet");
154+
model.summary();
155+
// compile keras model in tensorflow static graph
156+
model.compile(optimizer: keras.optimizers.RMSprop(1e-3f),
157+
loss: keras.losses.CategoricalCrossentropy(from_logits: true),
158+
metrics: new[] { "acc" });
159+
// prepare dataset
160+
var ((x_train, y_train), (x_test, y_test)) = keras.datasets.cifar10.load_data();
161+
x_train = x_train / 255.0f;
162+
y_train = np_utils.to_categorical(y_train, 10);
163+
// training
164+
model.fit(x_train[new Slice(0, 2000)], y_train[new Slice(0, 2000)],
165+
batch_size: 64,
166+
epochs: 10,
167+
validation_split: 0.2f);
168+
```
169+
170+
此外，Tensorflow.NET也支持用F#搭建上述模型进行训练和推理。
171+
172+
### Tensorflow.NET版本对应关系
173+
174+
| TensorFlow.NET Versions | tensorflow 1.14, cuda 10.0 | tensorflow 1.15, cuda 10.0 | tensorflow 2.3, cuda 10.1 | tensorflow 2.4, cuda 11 | tensorflow 2.10, cuda 11 |
175+
| -------------------------- | ------------- | -------------- | ------------- | ------------- | ------------ |
176+
| tf.net 0.7+, tf.keras 0.7+ | | | | | x |
177+
| tf.net 0.4x, tf.keras 0.5 | | | | x | |
178+
| tf.net 0.3x, tf.keras 0.4 | | | x | | |
179+
| tf.net 0.2x | | x | x | | |
180+
| tf.net 0.15 | x | x | | | |
181+
| tf.net 0.14 | x | | | | |
182+
183+
如果使用过程中发现有缺失的版本，请告知我们，谢谢！
184+
185+
请注意Tensorflow.NET与Tensorflow.Keras版本存在一一对应关系，请安装与Tensorflow.NET对应的Tensorflow.Keras版本。
186+
187+
### 参与我们的开发:
188+
189+
我们欢迎任何人的任何形式的贡献！无论是文档中的错误纠正，新特性提议，还是BUG修复等等，都会使得Tensorflow.NET项目越来越好，Tensorflow.NET的全体开发者也会积极帮助解决您提出的问题。
190+
191+
下面任何一种形式都可以帮助Tensorflow.NET越来越好：
192+
193+
* Star和分享Tensorflow.NET项目
194+
* 为Tensorflow.NET添加更多的用例
195+
* 在issue中告知我们Tensorflow.NET目前相比tensorflow缺少的API或者没有对齐的特性
196+
* 在issue中提出Tensorflow.NET存在的BUG或者可以改进的地方
197+
* 在待办事项清单中选择一个进行或者解决某个issue
198+
* 帮助我们完善文档，这也十分重要
199+
200+
201+
### 支持我们
202+
我们推出了[TensorFlow.NET实战](https://item.jd.com/13441549.html)这本书，包含了Tensorflow.NET主要开发者编写的讲解与实战例程，欢迎您的购买，希望这本书可以给您带来帮助。
203+
<p float="left">
204+
<img src="https://user-images.githubusercontent.com/1705364/198852429-91741881-c196-401e-8e9e-2f8656196613.png" width="250" />
205+
<img src="https://user-images.githubusercontent.com/1705364/198852521-2f842043-3ace-49d2-8533-039c6a043a3f.png" width="260" />
206+
<img src="https://user-images.githubusercontent.com/1705364/198852721-54cd9e7e-9210-4931-a86c-77584b25b8e1.png" width="260" />
207+
</p>
208+
209+
### 联系我们
210+
211+
可以在 [Twitter](https://twitter.com/ScisharpStack), [Facebook](https://www.facebook.com/scisharp.stack.9), [Medium](https://medium.com/scisharp), [LinkedIn](https://www.linkedin.com/company/scisharp-stack/)中关注我们，也可以在[Gitter](https://gitter.im/sci-sharp/community)中与项目开发者以及其它使用者进行沟通交流，也欢迎在仓库中提起issue。
212+
213+
TensorFlow.NET is a part of [SciSharp STACK](https://scisharp.github.io/SciSharp/)
214+
<br>
215+
<a href="http://scisharpstack.org"><img src="https://github.com/SciSharp/SciSharp/blob/master/art/scisharp-stack.png" width="391" height="100" /></a>

0 commit comments

Comments
 (0)