tensorflow入门

1、TensorFlow express a numeric computation as a graph

Graph nodes are operations which have any number of inputs and outputs

Graph edges are tensors which flow between nodes

tensorflow计算图

2、Variables, Placeholders & Operations

（1）Variables are stateful nodes which output their current value. State is retained across multiple executions of a graph (mostly parameters)

（2）Placeholders are nodes whose value is fed in at execution time (inputs, labels, …)

（3） Mathematical operations:

① MatMul: Multiply two matrix values.

② Add: Add elementwise (with broadcasting).

③ ReLU: Activate with elementwise rectified linear function.

3、Steps of writing and running programs in TensorFlow

（1）创建tensors (variables)

（2）定义tensors之间的操作

（3）初始化tensors

（4）创建会话

（5）运行会话 

示例代码如下：

---

》First build a graph using variables and placeholders:

创建计算图

》Then deploy the graph onto a session, which is the execution environment:

执行计算图

4、How to train the model ？

（1）Build a graph  创建计算图

（2）Feedforward / Prediction  前向传播 / 预测

（3）Optimization (gradients and train_step operation)  优化

（4）Initialize a session  初始化会话

（5）Train with session.run(train_step, feed_dict)  运行会话，训练模型

代码示例如下：

---

定义损失 计算梯度 构建训练步骤 训练模型

5、Variables sharing

变量共享

---

参考资料：斯坦福CS224N深度学习自然语言处理课程