Welcome to advisor’s documentation!¶

Advisor¶

简介¶

Advisor is the hyper parameters tuning system for black box optimization.

It is the open-source implementation of Google Vizier with these features.

Easy to use with API, SDK, WEB and CLI
Support abstractions of Study and Trial
Included search and early stop algorithms
Recommend parameters with trained model
Same programming interfaces as Google Vizier
Command-line tool just like Microsoft NNI.

算法¶

[x] Grid Search
[x] Random Search
[x] Bayesian Optimization
[x] TPE(Hyperopt)
[x] Random Search(Hyperopt)
[x] Simulate Anneal(Hyperopt)
[x] Quasi Random(Chocolate)
[x] Grid Search(Chocolate)
[x] Random Search(Chocolate)
[x] Bayes(Chocolate)
[x] CMAES(Chocolate)
[x] MOCMAES(Chocolate)
[ ] SMAC Algorithm
[x] Early Stop First Trial Algorithm
[x] Early Stop Descending Algorithm
[ ] Performance Curve Stop Algorithm

安装¶

Pip¶

pip install advisor

From Source¶

git clone git@github.com:tobegit3hub/advisor.git

cd ./advisor/advisor_client/

python ./setup.py install

Docker¶

docker run -d -p 8000:8000 tobegit3hub/advisor

Docker Compose¶

wget https://raw.githubusercontent.com/tobegit3hub/advisor/master/docker-compose.yml

docker-compose up -d

Kubernetes¶

wget https://raw.githubusercontent.com/tobegit3hub/advisor/master/kubernetes_advisor.yaml

kubectl create -f ./kubernetes_advisor.yaml

快速使用¶

使用pip安装。

pip install advisor

启动服务器。

advisor_admin server start

在浏览器打开http://127.0.0.1:8000。

提交调优任务。

git clone --depth 1 https://github.com/tobegit3hub/advisor.git && cd ./advisor/

advisor run -f ./advisor_client/examples/python_function/config.json

获取任务结果。

advisor study describe -s demo

服务器¶

命令行¶

advisor_admin server start

Docker¶

docker run -d -p 8000:8000 tobegit3hub/advisor

Docker Compose¶

wget https://raw.githubusercontent.com/tobegit3hub/advisor/master/docker-compose.yml

docker-compose up -d

Kubernetes¶

wget https://raw.githubusercontent.com/tobegit3hub/advisor/master/kubernetes_advisor.yaml

kubectl create -f ./kubernetes_advisor.yaml

源码启动¶

git clone --depth 1 https://github.com/tobegit3hub/advisor.git && cd ./advisor/

pip install -r ./requirements.txt

./manage.py migrate

./manage.py runserver 0.0.0.0:8000

命令行¶

启动服务¶

advisor_admin server start

停止服务¶

advisor_admin server stop

提交任务¶

advisor run -f ./advisor_client/examples/python_function/config.json

列举Study¶

advisor study list

查看Study¶

advisor study describe -s demo

列举Trial¶

advisor trials list

SDK¶

创建客户端对象¶

client = AdvisorClient()

创建Study¶

study_configuration = {
  "goal":
  "MINIMIZE",
  "randomInitTrials":
  1,
  "maxTrials":
  5,
  "maxParallelTrials":
  1,
  "params": [
      {
          "parameterName": "gamma",
          "type": "DOUBLE",
          "minValue": 0.001,
          "maxValue": 0.01,
          "feasiblePoints": "",
          "scalingType": "LINEAR"
      },
      {
          "parameterName": "C",
          "type": "DOUBLE",
          "minValue": 0.5,
          "maxValue": 1.0,
          "feasiblePoints": "",
          "scalingType": "LINEAR"
      },
      {
          "parameterName": "kernel",
          "type": "CATEGORICAL",
          "minValue": 0,
          "maxValue": 0,
          "feasiblePoints": "linear, poly, rbf, sigmoid, precomputed",
          "scalingType": "LINEAR"
      },
      {
          "parameterName": "coef0",
          "type": "DOUBLE",
          "minValue": 0.0,
          "maxValue": 0.5,
          "feasiblePoints": "",
          "scalingType": "LINEAR"
      },
  ]
}
study = client.create_study("Study", study_configuration,
                          "BayesianOptimization")

获取Study¶

study = client.get_study_by_id(6)

获取Trial¶

trials = client.get_suggestions(study.id, 3)

生成参数¶

parameter_value_dicts = []
for trial in trials:
  parameter_value_dict = json.loads(trial.parameter_values)
  print("The suggested parameters: {}".format(parameter_value_dict))
  parameter_value_dicts.append(parameter_value_dict)

运行训练¶

metrics = []
for i in range(len(trials)):
  metric = train_function(**parameter_value_dicts[i])
  metrics.append(metric)

完成Trial¶

for i in range(len(trials)):
  trial = trials[i]
  client.complete_trial_with_one_metric(trial, metrics[i])
is_done = client.is_study_done(study.id)
best_trial = client.get_best_trial(study.id)
print("The study: {}, best trial: {}".format(study, best_trial))

配置文件¶

YAML示例¶

name: "demo"
algorithm: "BayesianOptimization"
trialNumber: 10
path: "./advisor_client/examples/python_function/"
command: "./min_function.py"
search_space:
  goal: "MINIMIZE"
  randomInitTrials: 3
  params:
    - parameterName: "x"
      type: "DOUBLE"
      minValue: -10.0
      maxValue: 10.0

JSON示例¶

{
  "name": "demo",
  "algorithm": "BayesianOptimization",
  "trialNumber": 10,
  "concurrency": 1,
  "path": "./advisor_client/examples/python_function/",
  "command": "./min_function.py",
  "search_space": {
      "goal": "MINIMIZE",
      "randomInitTrials": 3,
      "params": [
          {
              "parameterName": "x",
              "type": "DOUBLE",
              "minValue": -10.0,
              "maxValue": 10.0,
              "scalingType": "LINEAR"
          }
      ]
  }
}