Program
class paddle.static.Program
[源代码]
注意:默认情况下,Paddle内部默认含有 default_startup_program 和 default_main_program ,它们共享参数。 default_startup_program 只运行一次来初始化参数, default_main_program 在每个mini batch中运行并更新权重。
Program是Paddle对于计算图的一种静态描述,使用Program的构造函数可以创建一个Program。Program中包括至少一个 Block ,当 Block 中存在条件选择的控制流OP(例如 While 等)时,该Program将会含有嵌套着的 Block 即控制流外部的 Block 将包含着控制流内部的 Block ,而嵌套的 Block 的元素访问控制将由具体的控制流OP来决定。关于Program具体的结构和包含的类型请参阅 framework.proto 。
一个Program的集合通常包含初始化程序(startup_program)与主程序(main_program),初始化程序是一个包含一些初始化工作的Program,主程序将会包含用来训练的网络结构和变量,在使用同一个 执行引擎 执行时他们会共享初始化工作的结果,例如初始化的参数。一个Program的集合可以被用来测试或者训练,被用来训练时, Paddle
将会利用所有用户使用的OP和变量来搭建一个训练网络,被用来测试时, 可以通过调用Program相关的接口例如:clone 剪去一些与测试无关的OP和变量,比如反向传播的OP和变量。
返回
Program,创建的空的Program
代码示例
import paddle
import paddle.static as static
paddle.enable_static()
main_program = static.Program()
startup_program = static.Program()
with static.program_guard(main_program=main_program, startup_program=startup_program):
x = static.data(name="x", shape=[-1, 784], dtype='float32')
y = static.data(name="y", shape=[-1, 1], dtype='int32')
z = static.nn.fc(name="fc", x=x, size=10, activation="relu")
print("main program is: {}".format(main_program))
print("start up program is: {}".format(startup_program))
to_string
( throw_on_error, with_details=False )
将Program转换为字符串
参数
throw_on_error (bool) - 是否在没有设置必需字段时抛出异常。
with_details (bool) - 值为true时,打印更多关于变量和参数的信息,如trainable, optimize_attr等
返回
str,由Program转换得到的字符串
抛出异常: ValueError
- 当 throw_on_error == true
,当没有设置任何必需的字段时,抛出 ValueError
。
代码示例
import paddle
import paddle.static as static
paddle.enable_static()
prog = static.default_main_program()
x = static.data(name="X", shape=[2,3], dtype="float32")
pred = static.nn.fc(x, size=3)
prog_string = prog.to_string(throw_on_error=True, with_details=False)
prog_string_with_details = prog.to_string(throw_on_error=False, with_details=True)
print("program string without detail: {}".format(prog_string))
print("program string with detail: {}".format(prog_string_with_details))
clone
( for_test=False )
注解
Program.clone()
方法不会克隆例如 DataLoader 这样的数据读取相关的部分,这可能会造成的数据读取部分在克隆后丢失;此API当
for_test=True
时将会裁剪部分OP和变量。为防止错误的裁剪,推荐在 append_backward 和执行优化器之前使用;clone(for_test=True)
。
当 for_test=True
时创建一个新的、仅包含当前Program前向内容的Program。否则创建一个新的,和当前Program完全相同的Program
有些OP,在训练和测试之间的行为是不同的,比如 batch_norm 。它们有一个属性 is_test
来控制行为。当 for_test=True
时,此方法将把它们的 is_test
属性更改为True。
克隆Program用于训练时,将
for_test
设置为False。克隆Program用于测试时,将
for_test
设置为True。虽然在这种情况下,如果在使用了优化器之后调用clone
我们依旧会对Program当中反向执行以及优化器相关的内容进行自动裁剪,但是,我们强烈建议在使用优化器之前使用clone
例如如果使用的是 Momentum 可以这样去使用:
代码示例
import paddle
import paddle.static as static
paddle.enable_static()
img = static.data(name='image', shape=[None, 784])
pred = static.nn.fc(x=img, size=10, activation='relu')
loss = paddle.mean(pred)
# Here we use clone before Momentum
test_program = static.default_main_program().clone(for_test=True)
optimizer = paddle.optimizer.Momentum(learning_rate=0.01, momentum=0.9)
optimizer.minimize(loss)
参数
- for_test (bool) – 取值为True时,clone方法内部会把operator的属性
is_test
设置为 True, 并裁剪反向OP和参数优化OP,默认值为False
返回
Program,当 for_test=True
时返回一个新的、仅包含当前Program前向内容的Program。否则返回一个新的,和当前Program完全相同的Program
代码示例
注解
Program在clone后的顺序可能不同,这不会影响的训练或测试进程。在下面的示例中,我们提供了一个简单的方法print_prog(Program)来打印程序描述,以确保clone后仍能得到同样的打印结果:
import six
def print_prog(prog):
for name, value in sorted(six.iteritems(prog.block(0).vars)):
print(value)
for op in prog.block(0).ops:
print("op type is {}".format(op.type))
print("op inputs are {}".format(op.input_arg_names))
print("op outputs are {}".format(op.output_arg_names))
for key, value in sorted(six.iteritems(op.all_attrs())):
if key not in ['op_callstack', 'op_role_var']:
print(" [ attrs: {}: {} ]".format(key, value))
1.克隆一个Program,示例代码如下。
import six
import paddle
import paddle.static as static
import paddle.utils as utils
import paddle.nn.functional as F
paddle.enable_static()
def print_prog(prog):
for name, value in sorted(six.iteritems(prog.block(0).vars)):
print(value)
for op in prog.block(0).ops:
print("op type is {}".format(op.type))
print("op inputs are {}".format(op.input_arg_names))
print("op outputs are {}".format(op.output_arg_names))
for key, value in sorted(six.iteritems(op.all_attrs())):
if key not in ['op_callstack', 'op_role_var']:
print(" [ attrs: {}: {} ]".format(key, value))
train_program = static.Program()
startup_program = static.Program()
# startup_program is used to do some parameter init work,
# and main program is used to hold the network
with static.program_guard(train_program, startup_program):
with utils.unique_name.guard():
img = static.data(name='image', shape=[None, 784])
hidden = static.nn.fc(x=img, size=200, activation='relu')
hidden = F.dropout(hidden, p=0.5)
loss = F.cross_entropy(
input=static.nn.fc(x=hidden, size=10, activation='softmax'),
label=static.data(name='label', shape=[1], dtype='int64'))
avg_loss = paddle.mean(loss)
test_program = train_program.clone(for_test=True)
print_prog(test_program)
# Due to parameter sharing usage for train and test, so we need to use startup program of train
# instead of using test startup program, while nothing is in test's startup program
# In Paddle we will share weights by using the same Tensor name. In train and test program
# all parameters will have the same name and this can make train and test program sharing parameters,
# that's why we need to use startup program of train. And for startup program of test, it has nothing,
# since it is a new program.
with static.program_guard(train_program, startup_program):
with utils.unique_name.guard():
sgd = paddle.optimizer.SGD(learning_rate=1e-3)
sgd.minimize(avg_loss)
2.如果分别运行 train Program 和 test Program,则可以不使用clone。
import six
import paddle
import paddle.static as static
import paddle.utils as utils
import paddle.nn.functional as F
paddle.enable_static()
def print_prog(prog):
for name, value in sorted(six.iteritems(prog.block(0).vars)):
print(value)
for op in prog.block(0).ops:
print("op type is {}".format(op.type))
print("op inputs are {}".format(op.input_arg_names))
print("op outputs are {}".format(op.output_arg_names))
for key, value in sorted(six.iteritems(op.all_attrs())):
if key not in ['op_callstack', 'op_role_var']:
print(" [ attrs: {}: {} ]".format(key, value))
def network():
img = static.data(name='image', shape=[None, 784])
hidden = static.nn.fc(x=img, size=200, activation='relu')
hidden = F.dropout(hidden, p=0.5)
loss = F.cross_entropy(
input=static.nn.fc(x=hidden, size=10, activation='softmax'),
label=static.data(name='label', shape=[1], dtype='int64'))
avg_loss = paddle.mean(loss)
return avg_loss
train_program_2 = static.Program()
startup_program_2 = static.Program()
test_program_2 = static.Program()
with static.program_guard(train_program_2, startup_program_2):
with utils.unique_name.guard():
avg_loss = network()
sgd = paddle.optimizer.SGD(learning_rate=1e-3)
sgd.minimize(avg_loss)
# the test startup program is not used.
with static.program_guard(test_program_2, startup_program_2):
with utils.unique_name.guard():
avg_loss = network()
print_prog(test_program_2)
上边两个代码片段生成和打印的Program是一样的。
static parse_from_string
( binary_str )
通过对 protobuf 的反序列化,转换成Program
参数
- binary_str_type (str) – protobuf 二进制字符串
返回
Program,反序列化后的 Program
代码示例
import paddle
import paddle.static as static
paddle.enable_static()
startup_prog = static.Program()
main_prog = static.Program()
with static.program_guard(startup_prog, main_prog):
x = static.data(name='X', shape=[1000, 784], dtype='float32')
y = static.data(name='Y', shape=[784, 100], dtype='float32')
z = paddle.matmul(x=x, y=y)
binary_str = static.default_main_program().desc.serialize_to_string()
prog_restored = static.default_main_program().parse_from_string(binary_str)
print(static.default_main_program())
print(prog_restored)
num_blocks
该Program中的 Block 的个数
返回
int,该Program中的 Block 的个数
代码示例
import paddle
import paddle.static as static
paddle.enable_static()
prog = static.default_main_program()
num_blocks = prog.num_blocks
print(num_blocks)
# print result:
# 1
random_seed
注解
必须在相关OP被添加之前设置。
程序中随机运算符的默认随机种子。0意味着随机生成随机种子。
返回
int64,该Program中当前正在使用的random seed
代码示例
import paddle
import paddle.static as static
import paddle.nn.functional as F
paddle.enable_static()
prog = static.default_main_program()
random_seed = prog.random_seed
x_var = static.data(name="X", shape=[3,3], dtype="float32")
print(random_seed)
## 0
## the default random seed is 0
# Here we need to set random seed before we use paddle.nn.functional.dropout
prog.random_seed = 1
z_var = F.dropout(x_var, 0.7)
print(prog.random_seed)
## 1
## the random seed is change to 1
global_block
( )
获取该Program的第一个 Block 。
返回
代码示例
import paddle
import paddle.static as static
paddle.enable_static()
prog = static.default_main_program()
gb_block = prog.global_block()
print(gb_block)
block
( index )
返回该Program中 , index
指定的 Block 。 index
类型为int
参数
- index (int) - 需要获取的 Block 的index
返回
Block,该Program中index对应的那个 Block
代码示例
import paddle
import paddle.static as static
paddle.enable_static()
prog = static.default_main_program()
block_0 = prog.block(0)
print(block_0)
current_block
( )
获取当前 Block 。当前 Block 是用来添加OP的。
返回
代码示例
import paddle
import paddle.static as static
paddle.enable_static()
prog = static.default_main_program()
current_blk = prog.current_block()
print(current_blk)
list_vars
( )
获取当前Program中所有变量。返回值是一个可迭代对象(iterable object)。
返回
Generator,会yield每个Program中的变量
代码示例
import paddle
import paddle.static as static
paddle.enable_static()
prog = static.default_main_program()
img = static.data(name='img', shape=[None, 1,28,28], dtype='float32')
label = static.data(name='label', shape=[None,1], dtype='int64')
for var in prog.list_vars():
print(var)
# var img : paddle.VarType.LOD_TENSOR.shape(-1, 1, 28, 28).astype(VarType.FP32)
# var label : paddle.VarType.LOD_TENSOR.shape(-1, 1).astype(VarType.INT64)
all_parameters
( )
获取当前Program中所有的 模型参数 。返回值是一个列表。
返回
list[ 模型参数 ],一个包含当前Program中所有参数的列表。
代码示例
import paddle
import paddle.static as static
paddle.enable_static()
program = static.default_main_program()
data = static.data(name='x', shape=[None, 13], dtype='float32')
hidden = static.nn.fc(x=data, size=10)
loss = paddle.mean(hidden)
paddle.optimizer.SGD(learning_rate=0.01).minimize(loss)
for param in program.all_parameters():
print(param)
# Here will print all parameters in current program, in this example,
# the result is like:
#
# persist trainable param fc_0.w_0 : paddle.VarType.LOD_TENSOR.shape(13, 10).astype(VarType.FP32)
# persist trainable param fc_0.b_0 : paddle.VarType.LOD_TENSOR.shape(10,).astype(VarType.FP32)
#
# Here print(param) will print out all the properties of a parameter,
# including name, type and persistable, you can access to specific
# property of a parameter, such as param.name, param.type