Polish examples in PR 4210

[wangkuiyi]

Polish examples in https://github.com/PaddlePaddle/Paddle/pull/4210/files

ifelse op

import paddle as pd

x = minibatch([10, 20, 30]) # shape=[None, 1]
y = var(1) # shape=[1], value=1
z = minibatch([10, 20, 30]) # shape=[None, 1]
cond = larger_than(x, 15) # [false, true, true]

ie = pd.ifelse()
with ie.true_block():
    d = pd.layer.add_scalar(x, y)
    ie.output(d, pd.layer.softmax(d))
with ie.false_block():
    d = pd.layer.fc(z)
    ie.output(d, d+1)
o1, o2 = ie(cond)

namespace pd = paddle;

int x = 10;
int y = 1;
int z = 10;
bool cond = false;
int o1, o2;
if (cond) {
  int d = x + y;
  o1 = z;
  o2 = pd::layer::softmax(z);
} else {
  int d = pd::layer::fc(z);
  o1 = d;
  o2 = d+1;
}

[jacquesqiao]

with ie.true_block() as b:
    pass

[Superjomn]

python supports the syntax like with ie.true_block(): pass @jacquesqiao

[Superjomn]

rnn op

x = sequence([10, 20, 30]) # shape=[None, 1]
m = var(0) # shape=[1]
W = var(0.314, param=true) # shape=[1]
U = var(0.375, param=true) # shape=[1]

rnn = pd.rnn()
with rnn.step():
  x_ = rnn.step_input(x)
  h = rnn.memory(init = m)
  hh = rnn.previous_memory(h)
  a = layer.fc(W, x_)
  b = layer.fc(U, hh)  
  s = pd.add(a, b)
  act = pd.sigmoid(s)
  rnn.update_memory(h, act)
  rnn.output(a, b)
o1, o2 = rnn()

has its equivalent C++ program as follows

int* x = {10, 20, 30};
int* m = {0};
int* W = {0.314};
int* U = {0.375};

int mem[sizeof(x) / sizeof(x[0]) + 1];
int o1[sizeof(x) / sizeof(x[0]) + 1];
int o2[sizeof(x) / sizeof(x[0]) + 1];
for (int i = 1; i <= sizeof(x)/sizeof(x[0]); ++i) {
  int x = x[i-1];
  if (i == 1) mem[0] = m;
  int a = W * x;
  int b = Y * mem[i-1];
  int s = fc_out + hidden_out;
  int act = sigmoid(sum);
  mem[i] = act;
  o1[i] = act;
  o2[i] = hidden_out;
}

[Superjomn]

with-block warning when local variables defined in rnn-step-block is used outside, may need to add some info in Variable. @wangkuiyi

[Superjomn]

x_ = rnn.step_input(x)

if we can add some block-related-info in Variable, why not x_ = x.as_step_input() this is more concise. @wangkuiyi

[Superjomn]

another style of RNNOp's syntax may be

x = sequence([10, 20, 30]) # shape=[None, 1]
m = var(0) # shape=[1]
W = var(0.314, param=true) # shape=[1]
U = var(0.375, param=true) # shape=[1]

rnn = pd.rnn()
with rnn.step():
  x_ = x.as_step_input()
  h = rnn.memory(init = m)
  hh = h.prev();
  a = layer.fc(W, x_)
  b = layer.fc(U, hh)  
  s = pd.add(a, b)
  act = pd.sigmoid(s)
  h.update(act)
  rnn.output(a, b)
o1, o2 = rnn()

[wangkuiyi]

I need to update examples here into if_else_op.md and rnn.md.