Skip to content
体验新版
项目
组织
正在加载...
登录
切换导航
打开侧边栏
一只大灰猫
PaddleRec
提交
e8d99992
P
PaddleRec
项目概览
一只大灰猫
/
PaddleRec
与 Fork 源项目一致
Fork自
PaddlePaddle / PaddleRec
通知
1
Star
1
Fork
0
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
0
列表
看板
标记
里程碑
合并请求
0
Wiki
0
Wiki
分析
仓库
DevOps
项目成员
Pages
P
PaddleRec
项目概览
项目概览
详情
发布
仓库
仓库
文件
提交
分支
标签
贡献者
分支图
比较
Issue
0
Issue
0
列表
看板
标记
里程碑
合并请求
0
合并请求
0
Pages
分析
分析
仓库分析
DevOps
Wiki
0
Wiki
成员
成员
收起侧边栏
关闭侧边栏
动态
分支图
创建新Issue
提交
Issue看板
体验新版 GitCode,发现更多精彩内容 >>
提交
e8d99992
编写于
5月 27, 2020
作者:
X
xjqbest
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
fix
上级
e91f9dd2
变更
5
隐藏空白更改
内联
并排
Showing
5 changed file
with
232 addition
and
7 deletion
+232
-7
core/factory.py
core/factory.py
+2
-0
core/trainers/single_auc_yamlopt.py
core/trainers/single_auc_yamlopt.py
+214
-0
core/trainers/single_trainer_yamlopt.py
core/trainers/single_trainer_yamlopt.py
+0
-6
models/rank/dnn/config.yaml
models/rank/dnn/config.yaml
+1
-0
run.py
run.py
+15
-1
未找到文件。
core/factory.py
浏览文件 @
e8d99992
...
...
@@ -38,6 +38,8 @@ def trainer_registry():
"tdm_cluster_trainer.py"
)
trainers
[
"SingleTrainerYamlOpt"
]
=
os
.
path
.
join
(
trainer_abs
,
"single_trainer_yamlopt.py"
)
trainers
[
"SingleAucYamlOpt"
]
=
os
.
path
.
join
(
trainer_abs
,
"single_auc_yamlopt.py"
)
trainer_registry
()
...
...
core/trainers/single_auc_yamlopt.py
0 → 100755
浏览文件 @
e8d99992
# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""
Training use fluid with one node only.
"""
from
__future__
import
print_function
import
time
import
logging
import
os
import
paddle.fluid
as
fluid
from
paddlerec.core.trainers.transpiler_trainer
import
TranspileTrainer
from
paddlerec.core.utils
import
envs
from
paddlerec.core.reader
import
SlotReader
logging
.
basicConfig
(
format
=
"%(asctime)s - %(levelname)s - %(message)s"
)
logger
=
logging
.
getLogger
(
"fluid"
)
logger
.
setLevel
(
logging
.
INFO
)
class
SingleAucYamlOpt
(
TranspileTrainer
):
def
__init__
(
self
,
config
=
None
):
super
(
TranspileTrainer
,
self
).
__init__
(
config
)
self
.
_env
=
self
.
_config
device
=
envs
.
get_global_env
(
"device"
)
if
device
==
'gpu'
:
self
.
_place
=
fluid
.
CUDAPlace
(
0
)
elif
device
==
'cpu'
:
self
.
_place
=
fluid
.
CPUPlace
()
self
.
_exe
=
fluid
.
Executor
(
self
.
_place
)
self
.
processor_register
()
self
.
_model
=
{}
self
.
_dataset
=
{}
envs
.
set_global_envs
(
self
.
_config
)
envs
.
update_workspace
()
def
processor_register
(
self
):
self
.
regist_context_processor
(
'uninit'
,
self
.
instance
)
self
.
regist_context_processor
(
'init_pass'
,
self
.
init
)
self
.
regist_context_processor
(
'startup_pass'
,
self
.
startup
)
self
.
regist_context_processor
(
'train_pass'
,
self
.
executor_train
)
self
.
regist_context_processor
(
'terminal_pass'
,
self
.
terminal
)
def
instance
(
self
,
context
):
context
[
'status'
]
=
'init_pass'
def
_create_dataset
(
self
,
dataset_name
):
name
=
"dataset."
+
dataset_name
+
"."
sparse_slots
=
envs
.
get_global_env
(
name
+
"sparse_slots"
)
dense_slots
=
envs
.
get_global_env
(
name
+
"dense_slots"
)
thread_num
=
envs
.
get_global_env
(
name
+
"thread_num"
)
batch_size
=
envs
.
get_global_env
(
name
+
"batch_size"
)
reader_type
=
envs
.
get_global_env
(
name
+
"type"
)
if
envs
.
get_platform
()
!=
"LINUX"
:
print
(
"platform "
,
envs
.
get_platform
(),
" change reader to DataLoader"
)
reader_type
=
"DataLoader"
padding
=
0
reader
=
envs
.
path_adapter
(
"paddlerec.core.utils"
)
+
"/dataset_instance.py"
pipe_cmd
=
"python {} {} {} {} {} {} {} {}"
.
format
(
reader
,
"slot"
,
"slot"
,
self
.
_config_yaml
,
"fake"
,
\
sparse_slots
.
replace
(
" "
,
"#"
),
dense_slots
.
replace
(
" "
,
"#"
),
str
(
padding
))
type_name
=
envs
.
get_global_env
(
name
+
"type"
)
if
type_name
==
"QueueDataset"
:
dataset
=
fluid
.
DatasetFactory
().
create_dataset
()
dataset
.
set_batch_size
(
envs
.
get_global_env
(
name
+
"batch_size"
))
dataset
.
set_pipe_command
(
pipe_cmd
)
train_data_path
=
envs
.
get_global_env
(
name
+
"data_path"
)
file_list
=
[
os
.
path
.
join
(
train_data_path
,
x
)
for
x
in
os
.
listdir
(
train_data_path
)
]
dataset
.
set_filelist
(
file_list
)
for
model_dict
in
self
.
_env
[
"executor"
]:
if
model_dict
[
"dataset_name"
]
==
dataset_name
:
model
=
self
.
_model
[
model_dict
[
"name"
]][
3
]
inputs
=
model
.
get_inputs
()
dataset
.
set_use_var
(
inputs
)
break
else
:
pass
return
dataset
def
init
(
self
,
context
):
for
model_dict
in
self
.
_env
[
"executor"
]:
self
.
_model
[
model_dict
[
"name"
]]
=
[
None
]
*
4
train_program
=
fluid
.
Program
()
startup_program
=
fluid
.
Program
()
scope
=
fluid
.
Scope
()
opt_name
=
envs
.
get_global_env
(
"hyper_parameters.optimizer.class"
)
opt_lr
=
envs
.
get_global_env
(
"hyper_parameters.optimizer.learning_rate"
)
opt_strategy
=
envs
.
get_global_env
(
"hyper_parameters.optimizer.strategy"
)
with
fluid
.
program_guard
(
train_program
,
startup_program
):
with
fluid
.
unique_name
.
guard
():
model_path
=
model_dict
[
"model"
].
replace
(
"{workspace}"
,
envs
.
path_adapter
(
self
.
_env
[
"workspace"
]))
model
=
envs
.
lazy_instance_by_fliename
(
model_path
,
"Model"
)(
self
.
_env
)
model
.
_data_var
=
model
.
input_data
(
dataset_name
=
model_dict
[
"dataset_name"
])
model
.
net
(
None
,
is_infer
=
True
)
optimizer
=
model
.
_build_optimizer
(
opt_name
,
opt_lr
,
opt_strategy
)
optimizer
.
minimize
(
model
.
_cost
)
self
.
_model
[
model_dict
[
"name"
]][
0
]
=
train_program
self
.
_model
[
model_dict
[
"name"
]][
1
]
=
startup_program
self
.
_model
[
model_dict
[
"name"
]][
2
]
=
scope
self
.
_model
[
model_dict
[
"name"
]][
3
]
=
model
for
dataset
in
self
.
_env
[
"dataset"
]:
self
.
_dataset
[
dataset
[
"name"
]]
=
self
.
_create_dataset
(
dataset
[
"name"
])
context
[
'status'
]
=
'startup_pass'
def
startup
(
self
,
context
):
for
model_dict
in
self
.
_env
[
"executor"
]:
with
fluid
.
scope_guard
(
self
.
_model
[
model_dict
[
"name"
]][
2
]):
self
.
_exe
.
run
(
self
.
_model
[
model_dict
[
"name"
]][
1
])
context
[
'status'
]
=
'train_pass'
def
executor_train
(
self
,
context
):
epochs
=
int
(
self
.
_env
[
"epochs"
])
for
j
in
range
(
epochs
):
for
model_dict
in
self
.
_env
[
"executor"
]:
reader_name
=
model_dict
[
"dataset_name"
]
name
=
"dataset."
+
reader_name
+
"."
begin_time
=
time
.
time
()
if
envs
.
get_global_env
(
name
+
"type"
)
==
"DataLoader"
:
self
.
_executor_dataloader_train
(
model_dict
)
else
:
self
.
_executor_dataset_train
(
model_dict
)
end_time
=
time
.
time
()
seconds
=
end_time
-
begin_time
print
(
"epoch {} done, time elasped: {}"
.
format
(
j
,
seconds
))
context
[
'status'
]
=
"terminal_pass"
def
_executor_dataset_train
(
self
,
model_dict
):
reader_name
=
model_dict
[
"dataset_name"
]
model_name
=
model_dict
[
"name"
]
model_class
=
self
.
_model
[
model_name
][
3
]
fetch_vars
=
[]
fetch_alias
=
[]
fetch_period
=
20
metrics
=
model_class
.
get_metrics
()
if
metrics
:
fetch_vars
=
metrics
.
values
()
fetch_alias
=
metrics
.
keys
()
scope
=
self
.
_model
[
model_name
][
2
]
program
=
self
.
_model
[
model_name
][
0
]
reader
=
self
.
_dataset
[
reader_name
]
with
fluid
.
scope_guard
(
scope
):
self
.
_exe
.
train_from_dataset
(
program
=
program
,
dataset
=
reader
,
fetch_list
=
fetch_vars
,
fetch_info
=
fetch_alias
,
print_period
=
fetch_period
)
def
_executor_dataloader_train
(
self
,
model_dict
):
reader_name
=
model_dict
[
"dataset_name"
]
model_name
=
model_dict
[
"name"
]
model_class
=
self
.
_model
[
model_name
][
3
]
self
.
_model
[
model_name
][
1
]
=
fluid
.
compiler
.
CompiledProgram
(
self
.
_model
[
model_name
][
1
]).
with_data_parallel
(
loss_name
=
model_class
.
get_avg_cost
().
name
)
fetch_vars
=
[]
fetch_alias
=
[]
fetch_period
=
20
metrics
=
model_class
.
get_metrics
()
if
metrics
:
fetch_vars
=
metrics
.
values
()
fetch_alias
=
metrics
.
keys
()
metrics_varnames
=
[]
metrics_format
=
[]
metrics_format
.
append
(
"{}: {{}}"
.
format
(
"epoch"
))
metrics_format
.
append
(
"{}: {{}}"
.
format
(
"batch"
))
for
name
,
var
in
model_class
.
items
():
metrics_varnames
.
append
(
var
.
name
)
metrics_format
.
append
(
"{}: {{}}"
.
format
(
name
))
metrics_format
=
", "
.
join
(
metrics_format
)
reader
=
self
.
_dataset
[
reader_name
]
reader
.
start
()
batch_id
=
0
scope
=
self
.
_model
[
model_name
][
2
]
prorgram
=
self
.
_model
[
model_name
][
0
]
with
fluid
.
scope_guard
(
scope
):
try
:
while
True
:
metrics_rets
=
self
.
_exe
.
run
(
program
=
program
,
fetch_list
=
metrics_varnames
)
metrics
=
[
epoch
,
batch_id
]
metrics
.
extend
(
metrics_rets
)
if
batch_id
%
self
.
fetch_period
==
0
and
batch_id
!=
0
:
print
(
metrics_format
.
format
(
*
metrics
))
batch_id
+=
1
except
fluid
.
core
.
EOFException
:
reader
.
reset
()
def
terminal
(
self
,
context
):
context
[
'is_exit'
]
=
True
core/trainers/single_trainer_yamlopt.py
浏览文件 @
e8d99992
...
...
@@ -57,12 +57,6 @@ class SingleTrainerYamlOpt(TranspileTrainer):
def
instance
(
self
,
context
):
context
[
'status'
]
=
'init_pass'
def
dataloader_train
(
self
,
context
):
pass
def
dataset_train
(
self
,
context
):
pass
def
_create_dataset
(
self
,
dataset_name
):
name
=
"dataset."
+
dataset_name
+
"."
sparse_slots
=
envs
.
get_global_env
(
name
+
"sparse_slots"
)
...
...
models/rank/dnn/config.yaml
浏览文件 @
e8d99992
...
...
@@ -39,6 +39,7 @@ hyper_parameters:
epoch
:
trainer_class
:
single_yamlopt
#trainer_class: single_auc_yamlopt
save_checkpoint_interval
:
2
save_inference_interval
:
4
save_checkpoint_path
:
"
increment"
...
...
run.py
浏览文件 @
e8d99992
...
...
@@ -28,7 +28,7 @@ device = ["CPU", "GPU"]
clusters
=
[
"SINGLE"
,
"LOCAL_CLUSTER"
,
"CLUSTER"
]
engine_choices
=
[
"SINGLE"
,
"LOCAL_CLUSTER"
,
"CLUSTER"
,
"TDM_SINGLE"
,
"TDM_LOCAL_CLUSTER"
,
"TDM_CLUSTER"
,
"SINGLE_YAMLOPT"
"TDM_CLUSTER"
,
"SINGLE_YAMLOPT"
,
"SINGLE_AUC_YAMLOPT"
]
custom_model
=
[
'TDM'
]
model_name
=
""
...
...
@@ -43,6 +43,7 @@ def engine_registry():
engines
[
"TRANSPILER"
][
"CLUSTER"
]
=
cluster_engine
engines
[
"TRANSPILER"
][
"SINGLE_YAMLOPT"
]
=
single_yamlopt_engine
engines
[
"TRANSPILER"
][
"SINGLE_AUC_YAMLOPT"
]
=
single_auc_yamlopt_engine
engines
[
"PSLIB"
][
"SINGLE"
]
=
local_mpi_engine
engines
[
"PSLIB"
][
"LOCAL_CLUSTER"
]
=
local_mpi_engine
...
...
@@ -151,6 +152,19 @@ def single_yamlopt_engine(args):
trainer
=
TrainerFactory
.
create
(
args
.
model
)
return
trainer
def
single_auc_yamlopt_engine
(
args
):
trainer
=
get_trainer_prefix
(
args
)
+
"SingleAucYamlOpt"
single_envs
=
{}
single_envs
[
"train.trainer.trainer"
]
=
trainer
single_envs
[
"train.trainer.threads"
]
=
"2"
single_envs
[
"train.trainer.engine"
]
=
"single_yamlopt"
single_envs
[
"train.trainer.platform"
]
=
envs
.
get_platform
()
print
(
"use {} engine to run model: {}"
.
format
(
trainer
,
args
.
model
))
set_runtime_envs
(
single_envs
,
args
.
model
)
trainer
=
TrainerFactory
.
create
(
args
.
model
)
return
trainer
def
cluster_engine
(
args
):
def
update_workspace
(
cluster_envs
):
workspace
=
cluster_envs
.
get
(
"engine_workspace"
,
None
)
...
...
编辑
预览
Markdown
is supported
0%
请重试
或
添加新附件
.
添加附件
取消
You are about to add
0
people
to the discussion. Proceed with caution.
先完成此消息的编辑!
取消
想要评论请
注册
或
登录