Skip to content

S
Serving

PaddlePaddle / Serving
大约 1 年 前同步成功

通知 186
Star 833
Fork 253
S
Serving
未验证 提交 7d68f1dc 编写于 作者: J Jiawei Wang 提交者： GitHub
浏览文件
操作

Merge branch 'develop' into develop

上级 9c3eab21 32c7be83
隐藏空白更改
内联 并排
Showing with 129 addition and 46 deletion
doc/ABTEST_IN_PADDLE_SERVING.md
浏览文件 @ 7d68f1dc
......@@ -16,32 +16,30 @@ sh get_data.sh
```
### Processing Data
Data processing needs to use the relevant library, please use pip to install
``` shell
pip install paddlepaddle
pip install paddle-serving-app
pip install Shapely
````
The following Python code will process the data `test_data/part-0` and write to the `processed.data` file.
You can directly run the following command to process the data.
[//file]:#process.py
``` python
from paddle_serving_app.reader import IMDBDataset
imdb_dataset = IMDBDataset()
imdb_dataset.load_resource('imdb.vocab')
with open('test_data/part-0') as fin:
with open('processed.data', 'w') as fout:
for line in fin:
word_ids, label = imdb_dataset.get_words_and_label(line)
fout.write("{};{}\n".format(','.join([str(x) for x in word_ids]), label[0]))
```
[python abtest_get_data.py](../python/examples/imdb/abtest_get_data.py)
The Python code in the file will process the data `test_data/part-0` and write to the `processed.data` file.
### Start Server
Here, we [use docker](https://github.com/PaddlePaddle/Serving/blob/develop/doc/RUN_IN_DOCKER.md) to start the server-side service.
Here, we [use docker](RUN_IN_DOCKER.md) to start the server-side service.
First, start the BOW server, which enables the `8000` port:
``` shell
docker run -dit -v $PWD/imdb_bow_model:/model -p 8000:8000 --name bow-server hub.baidubce.com/paddlepaddle/serving:latest
docker exec -it bow-server bash
pip install paddle-serving-server
docker run -dit -v $PWD/imdb_bow_model:/model -p 8000:8000 --name bow-server hub.baidubce.com/paddlepaddle/serving:latest /bin/bash
docker exec -it bow-server /bin/bash
pip install paddle-serving-server -i https://pypi.tuna.tsinghua.edu.cn/simple
pip install paddle-serving-client -i https://pypi.tuna.tsinghua.edu.cn/simple
python -m paddle_serving_server.serve --model model --port 8000 >std.log 2>err.log &
exit
```
......@@ -49,18 +47,25 @@ exit
Similarly, start the LSTM server, which enables the `9000` port:
```bash
docker run -dit -v $PWD/imdb_lstm_model:/model -p 9000:9000 --name lstm-server hub.baidubce.com/paddlepaddle/serving:latest
docker exec -it lstm-server bash
pip install paddle-serving-server
docker run -dit -v $PWD/imdb_lstm_model:/model -p 9000:9000 --name lstm-server hub.baidubce.com/paddlepaddle/serving:latest /bin/bash
docker exec -it lstm-server /bin/bash
pip install paddle-serving-server -i https://pypi.tuna.tsinghua.edu.cn/simple
pip install paddle-serving-client -i https://pypi.tuna.tsinghua.edu.cn/simple
python -m paddle_serving_server.serve --model model --port 9000 >std.log 2>err.log &
exit
```
### Start Client
Run the following Python code on the host computer to start client. Make sure that the host computer is installed with the `paddle-serving-client` package.
In order to simulate abtest condition, you can run the following Python code on the host to start the client, but you need to ensure that the host has the relevant environment, you can also run in the docker environment.
[//file]:#ab_client.py
Before running, use `pip install paddle-serving-client` to install the paddle-serving-client package.
You can directly use the following command to make abtest prediction.
[python abtest_client.py](../python/examples/imdb/abtest_client.py)
[//file]:#abtest_client.py
``` python
from paddle_serving_client import Client
......@@ -91,7 +96,7 @@ In the code, the function `client.add_variant(tag, clusters, variant_weight)` is
When making prediction on the client side, if the parameter `need_variant_tag=True` is specified, the response will contain the variant tag corresponding to the distribution flow.
### Expected Results
Due to different network conditions, the results of each prediction may be slightly different.
``` python
[lstm](total: 1867) acc: 0.490091055169
[bow](total: 217) acc: 0.73732718894
......
doc/ABTEST_IN_PADDLE_SERVING_CN.md
浏览文件 @ 7d68f1dc
......@@ -16,31 +16,29 @@ sh get_data.sh
```
### 处理数据
由于处理数据需要用到相关库，请使用pip进行安装
``` shell
pip install paddlepaddle
pip install paddle-serving-app
pip install Shapely
````
您可以直接运行下面的命令来处理数据。
下面Python代码将处理`test_data/part-0`的数据，写入`processed.data`文件中。
[python abtest_get_data.py](../python/examples/imdb/abtest_get_data.py)
```python
from paddle_serving_app.reader import IMDBDataset
imdb_dataset = IMDBDataset()
imdb_dataset.load_resource('imdb.vocab')
with open('test_data/part-0') as fin:
with open('processed.data', 'w') as fout:
for line in fin:
word_ids, label = imdb_dataset.get_words_and_label(line)
fout.write("{};{}\n".format(','.join([str(x) for x in word_ids]), label[0]))
```
文件中的Python代码将处理`test_data/part-0`的数据，并将处理后的数据生成并写入`processed.data`文件中。
### 启动Server端
这里采用[Docker方式](https://github.com/PaddlePaddle/Serving/blob/develop/doc/RUN_IN_DOCKER_CN.md)启动Server端服务。
这里采用[Docker方式](RUN_IN_DOCKER_CN.md)启动Server端服务。
首先启动BOW Server，该服务启用`8000`端口：
```bash
docker run -dit -v $PWD/imdb_bow_model:/model -p 8000:8000 --name bow-server hub.baidubce.com/paddlepaddle/serving:latest
docker exec -it bow-server bash
docker run -dit -v $PWD/imdb_bow_model:/model -p 8000:8000 --name bow-server hub.baidubce.com/paddlepaddle/serving:latest /bin/bash
docker exec -it bow-server /bin/bash
pip install paddle-serving-server -i https://pypi.tuna.tsinghua.edu.cn/simple
pip install paddle-serving-client -i https://pypi.tuna.tsinghua.edu.cn/simple
python -m paddle_serving_server.serve --model model --port 8000 >std.log 2>err.log &
exit
```
......@@ -48,19 +46,27 @@ exit
同理启动LSTM Server，该服务启用`9000`端口：
```bash
docker run -dit -v $PWD/imdb_lstm_model:/model -p 9000:9000 --name lstm-server hub.baidubce.com/paddlepaddle/serving:latest
docker exec -it lstm-server bash
docker run -dit -v $PWD/imdb_lstm_model:/model -p 9000:9000 --name lstm-server hub.baidubce.com/paddlepaddle/serving:latest /bin/bash
docker exec -it lstm-server /bin/bash
pip install paddle-serving-server -i https://pypi.tuna.tsinghua.edu.cn/simple
pip install paddle-serving-client -i https://pypi.tuna.tsinghua.edu.cn/simple
python -m paddle_serving_server.serve --model model --port 9000 >std.log 2>err.log &
exit
```
### 启动Client端
为了模拟ABTEST工况，您可以在宿主机运行下面Python代码启动Client端，但需确保宿主机具备相关环境，您也可以在docker环境下运行.
运行前使用`pip install paddle-serving-client`安装paddle-serving-client包。
在宿主机运行下面Python代码启动Client端，需要确保宿主机装好`paddle-serving-client`包。
您可以直接使用下面的命令，进行ABTEST预测。
[python abtest_client.py](../python/examples/imdb/abtest_client.py)
```python
from paddle_serving_client import Client
import numpy as np
client = Client()
client.load_client_config('imdb_bow_client_conf/serving_client_conf.prototxt')
......@@ -68,28 +74,32 @@ client.add_variant("bow", ["127.0.0.1:8000"], 10)
client.add_variant("lstm", ["127.0.0.1:9000"], 90)
client.connect()
print('please wait for about 10s')
with open('processed.data') as f:
cnt = {"bow": {'acc': 0, 'total': 0}, "lstm": {'acc': 0, 'total': 0}}
for line in f:
word_ids, label = line.split(';')
word_ids = [int(x) for x in word_ids.split(',')]
feed = {"words": word_ids}
word_len = len(word_ids)
feed = {
"words": np.array(word_ids).reshape(word_len, 1),
"words.lod": [0, word_len]
}
fetch = ["acc", "cost", "prediction"]
[fetch_map, tag] = client.predict(feed=feed, fetch=fetch, need_variant_tag=True)
[fetch_map, tag] = client.predict(feed=feed, fetch=fetch, need_variant_tag=True,batch=True)
if (float(fetch_map["prediction"][0][1]) - 0.5) * (float(label[0]) - 0.5) > 0:
cnt[tag]['acc'] += 1
cnt[tag]['total'] += 1
for tag, data in cnt.items():
print('[{}](total: {}) acc: {}'.format(tag, data['total'], float(data['acc']) / float(data['total'])))
print('[{}](total: {}) acc: {}'.format(tag, data['total'], float(data['acc'])/float(data['total']) ))
```
代码中，`client.add_variant(tag, clusters, variant_weight)`是为了添加一个标签为`tag`、流量权重为`variant_weight`的variant。在这个样例中，添加了一个标签为`bow`、流量权重为`10`的BOW variant，以及一个标签为`lstm`、流量权重为`90`的LSTM variant。Client端的流量会根据`10:90`的比例分发到两个variant。
Client端做预测时，若指定参数`need_variant_tag=True`，返回值则包含分发流量对应的variant标签。
### 预期结果
由于网络情况的不同，可能每次预测的结果略有差异。
``` bash
[lstm](total: 1867) acc: 0.490091055169
[bow](total: 217) acc: 0.73732718894
......
python/examples/bert/prepare_model.py
浏览文件 @ 7d68f1dc
......@@ -16,7 +16,9 @@ import paddlehub as hub
import paddle.fluid as fluid
import sys
import paddle_serving_client.io as serving_io
import paddle
paddle.enable_static()
model_name = "bert_chinese_L-12_H-768_A-12"
module = hub.Module(model_name)
inputs, outputs, program = module.context(
......
python/examples/imdb/abtest_client.py 0 → 100644
浏览文件 @ 7d68f1dc
# 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.
from paddle_serving_client import Client
import numpy as np
client = Client()
client.load_client_config('imdb_bow_client_conf/serving_client_conf.prototxt')
client.add_variant("bow", ["127.0.0.1:8000"], 10)
client.add_variant("lstm", ["127.0.0.1:9000"], 90)
client.connect()
print('please wait for about 10s')
with open('processed.data') as f:
cnt = {"bow": {'acc': 0, 'total': 0}, "lstm": {'acc': 0, 'total': 0}}
for line in f:
word_ids, label = line.split(';')
word_ids = [int(x) for x in word_ids.split(',')]
word_len = len(word_ids)
feed = {
"words": np.array(word_ids).reshape(word_len, 1),
"words.lod": [0, word_len]
}
fetch = ["acc", "cost", "prediction"]
[fetch_map, tag] = client.predict(feed=feed, fetch=fetch, need_variant_tag=True,batch=True)
if (float(fetch_map["prediction"][0][1]) - 0.5) * (float(label[0]) - 0.5) > 0:
cnt[tag]['acc'] += 1
cnt[tag]['total'] += 1
for tag, data in cnt.items():
print('[{}](total: {}) acc: {}'.format(tag, data['total'], float(data['acc'])/float(data['total']) ))
python/examples/imdb/abtest_get_data.py 0 → 100644
浏览文件 @ 7d68f1dc
# 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.
from paddle_serving_app.reader.imdb_reader import IMDBDataset
imdb_dataset = IMDBDataset()
imdb_dataset.load_resource('imdb.vocab')
with open('test_data/part-0') as fin:
with open('processed.data', 'w') as fout:
for line in fin:
word_ids, label = imdb_dataset.get_words_and_label(line)
fout.write("{};{}\n".format(','.join([str(x) for x in word_ids]), label[0]))
Markdown is supported
0% 或 .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑！
想要评论请 注册 或