%------------------------------------------------------
\begin{tikzpicture}
\setlength{\base}{0.9cm}
\tikzstyle{rnnnode} = [rounded corners=1pt,minimum height=0.5\base,minimum width=1\base,draw,inner sep=0pt,outer sep=0pt]
\tikzstyle{wordnode} = [font=\tiny]
% RNN translation model
\begin{scope}[local bounding box=RNNMT]
% RNN Encoder
\coordinate (eemb0) at (0,0);
\foreach \x [count=\y from 0] in {1,2,...,3}
\node[rnnnode,minimum height=0.5\base,fill=green!30!white,anchor=west] (eemb\x) at ([xshift=0.4\base]eemb\y.east) {\tiny{$e_x()$}};
\foreach \x in {1,2,...,3}
\node[rnnnode,fill=blue!30!white,anchor=south] (enc\x) at ([yshift=0.3\base]eemb\x.north) {};
\node[] (enclabel1) at (enc1) {\tiny{$\vectorn{h}_{m-2}$}};
\node[] (enclabel2) at (enc2) {\tiny{$\vectorn{h}_{m-1}$}};
\node[rnnnode,fill=purple!30!white] (enclabel3) at (enc3) {\tiny{$\vectorn{h}_{m}$}};
\node[wordnode,left=0.4\base of enc1] (init1) {$\cdots$};
\node[wordnode,left=0.4\base of eemb1] (init2) {$\cdots$};
\node[wordnode,below=0pt of eemb1] () {走};
\node[wordnode,below=0pt of eemb2] () {吗};
\node[wordnode,below=0pt of eemb3] () {$\langle$eos$\rangle$};
% RNN Decoder
\foreach \x in {1,2,...,3}
\node[rnnnode,minimum height=0.5\base,fill=green!30!white,anchor=south] (demb\x) at ([yshift=\base]enc\x.north) {\tiny{$e_y()$}};
\foreach \x in {1,2,...,3}
\node[rnnnode,fill=blue!30!white,anchor=south] (dec\x) at ([yshift=0.3\base]demb\x.north) {{\tiny{$\vectorn{s}_\x$}}};
\foreach \x in {1,2,...,3}
\node[rnnnode,minimum height=0.5\base,fill=red!30!white,anchor=south] (softmax\x) at ([yshift=0.3\base]dec\x.north) {\tiny{Softmax}};
\node[wordnode,right=0.4\base of demb3] (end1) {$\cdots$};
\node[wordnode,right=0.4\base of dec3] (end2) {$\cdots$};
\node[wordnode,right=0.4\base of softmax3] (end3) {$\cdots$};
% Decoder input words
\node[wordnode,below=0pt of demb1] (decwordin) {$\langle$sos$\rangle$};
\ExtractX{$(demb2.south)$}
\ExtractY{$(decwordin.base)$}
\node[wordnode,anchor=base] () at (\XCoord,\YCoord) {Do};
\ExtractX{$(demb3.south)$}
\ExtractY{$(decwordin.base)$}
\node[wordnode,anchor=base] () at (\XCoord,\YCoord) {you};
% Decoder output words
\node[wordnode,above=0pt of softmax1] (decwordout) {Do};
\ExtractX{$(softmax2.north)$}
\ExtractY{$(decwordout.base)$}
\node[wordnode,anchor=base] () at (\XCoord,\YCoord) {you};
\ExtractX{$(softmax3.north)$}
\ExtractY{$(decwordout.base)$}
\node[wordnode,anchor=base] () at (\XCoord,\YCoord) {know};
% Connections
\draw[-latex'] (init1.east) to (enc1.west);
\draw[-latex'] (dec3.east) to (end2.west);
\foreach \x in {1,2,...,3}
\draw[-latex'] (eemb\x) to (enc\x);
\foreach \x in {1,2,...,3}
\draw[-latex'] (demb\x) to (dec\x);
\foreach \x in {1,2,...,3}
\draw[-latex'] (dec\x.north) to (softmax\x.south);
\foreach \x [count=\y from 2] in {1,2}
{
\draw[-latex'] (enc\x.east) to (enc\y.west);
\draw[-latex'] (dec\x.east) to (dec\y.west);
}
\coordinate (bridge) at ([yshift=0.4\base]enc2.north west);
\draw[-latex'] (enc3.north) .. controls +(north:0.3\base) and +(east:\base) .. (bridge) .. controls +(west:2.7\base) and +(west:0.3\base) .. (dec1.west);
\end{scope}
\begin{scope}
\coordinate (start) at (5.8\base,0.3\base);
{
\node [anchor=south west] (one) at (start) {\scriptsize{$\begin{bmatrix} 0 \\ 0 \\ 0 \\ \vdots \\ 0 \\ {\color{ugreen} 1} \\ 0 \\ 0 \end{bmatrix}$}};
\node [anchor=south west,inner sep=0pt] (T) at ([yshift=-0.5em,xshift=-0.5em]one.north east) {\tiny{T}};
}
{
\node [draw=ugreen,fill=green!20!white,rounded corners=0.3em,minimum width=3.8cm,minimum height=0.9em,anchor=south west] (emb) at ([shift={(1.25cm,0.8cm)}]start) {};
}
\node [anchor=north] (w) at ([yshift=3pt]one.south) {\scriptsize{\color{ugreen} you}};
\node [anchor=north west] (words) at ([xshift=10pt]one.north east) {\scriptsize{$\begin{matrix} \langle\textrm{eos}\rangle \\ \langle\textrm{sos}\rangle \\ \textrm{Do} \\ \vdots \\ \textrm{know} \\ \textrm{you} \\ \textrm{?} \\ \textrm{have} \end{matrix}$}};
\node [anchor=north west] (mat) at ([xshift=-6pt]words.north east) {\scriptsize{$
\begin{bmatrix}
.1 & -4 & \cdots & 2 \\
5 & 2 & \cdots & .2 \\
2 & .1 & \cdots & .3 \\
\vdots & \vdots & \ddots & \vdots \\
0 & .8 & \cdots & 4 \\
-1 & -2 & \cdots & -3 \\
.7 & .5 & \cdots & 3 \\
-2 & .3 & \cdots & .1
\end{bmatrix}
$}};
\draw [decorate,decoration={brace,mirror}] ([shift={(6pt,2pt)}]mat.south west) to node [auto,swap,font=\scriptsize] {词嵌入矩阵} ([shift={(-6pt,2pt)}]mat.south east);
{
\draw [-latex'] ([xshift=-2pt,yshift=-0.65cm]one.east) to ([yshift=-0.65cm]words.west);
}
{
\draw [-latex'] (emb.east) -| ([yshift=0.4cm]mat.north east) node [pos=1,above] {\scriptsize{RNN输入}};
}
\draw [-latex'] ([yshift=-0.4cm]w.south) to ([yshift=2pt]w.south);
\node [anchor=north] (wlabel) at ([yshift=-0.6em]w.south) {\scriptsize{输入的单词}};
\node [draw=ugreen,densely dashed,thick,rounded corners=3pt,fit=(one) (words) (mat) (w)] (input) {};
\end{scope}
\draw [->,thick,densely dashed,ugreen] ([yshift=-0.2em]demb3.east) to [out=0,in=180] ([yshift=-1cm]input.west);
\end{tikzpicture}