深度学习总结：tensorflow和pytorch关于RNN的对比,tf.nn.dynamic_rnn,nn.LSTM

tensorflow和pytorch关于RNN的对比：
tf.nn.dynamic_rnn很难理解，他的意思只是用数据走一遍你搭建的RNN网络。
可以明显看出pytorch封装更高，更容易理解，动态图的优势。

## tensorflow
# RNN
# num_units=64代表h_t,c_t的维度
rnn_cell = tf.contrib.rnn.BasicLSTMCell(num_units=64)
# 这个累加的lstm_multi ，相当于pytorch里面的num_layers=3
stacked_rnn=[]
for i in range(3):
    stacked_rnn.append(rnn_cell )
lstm_multi = tf.contrib.rnn.MultiRNNCell(stacked_rnn)

# tf.nn.dynamic_rnn这个就是在吓唬你，就是把batch放入lstm_multi里面跑，跑完后的各个time-step的输出
# 和最后一步的h_t,c_t，相当于r_out, (h_n, h_c) = self.rnn(x, None)，只不过tensorflow需要提前建好图
outputs, (h_c, h_n) = tf.nn.dynamic_rnn(
    lstm_multi ,                   # cell you have chosen
    image,                      # input
    initial_state=None,         # the initial hidden state
    dtype=tf.float32,           # must given if set initial_state = None
    time_major=False,           # False: (batch, time step, input); True: (time step, batch, input)
)

## pytorch
# hidden_size
        self.rnn = nn.LSTM(         # if use nn.RNN(), it hardly learns
            input_size=INPUT_SIZE,
            hidden_size=64,         # rnn hidden unit
            num_layers=3,           # number of rnn layer
            batch_first=True,       # input & output will has batch size as 1s dimension. e.g. (batch, time_step, input_size)
        )
   
r_out, (h_n, h_c) = self.rnn(x, None)   # None represents zero initial hidden state