优化多个目标

Colab 徽标 在 Colab 中以笔记本的形式运行本教程 GitHub 徽标在 GitHub 上查看笔记本

本教程将演示 AI Platform Optimizer 多目标的优化。

目标

目标是 minimize 目标指标:y1 = r*sin(theta)

并同时 maximize 目标指标:y2 = r*cos(theta)

您将通过参数空间进行评估:

  • r 的取值范围是 [0,1]

  • theta 的取值范围是 [0, pi/2]

费用

本教程使用 Google Cloud 的以下收费组件:

  • AI Platform Training
  • Cloud Storage

了解 AI Platform Training 价格Cloud Storage 价格,并使用价格计算器根据您的预计使用情况来估算费用。

PIP 安装软件包和依赖项

安装未在笔记本环境中安装的其他依赖项。

  • 使用最新的主要正式版框架。
! pip install -U google-api-python-client
! pip install -U google-cloud
! pip install -U google-cloud-storage
! pip install -U requests
! pip install -U matplotlib

# Restart the kernel after pip installs
import IPython
app = IPython.Application.instance()
app.kernel.do_shutdown(True)

设置您的 Google Cloud 项目

无论您使用哪种笔记本环境,都必须执行以下步骤。

  1. 选择或创建 Google Cloud 项目。

  2. 确保您的项目已启用结算功能。

  3. 启用 AI Platform API

  4. 如果在您自己的机器上本地运行,则需要安装 Google Cloud SDK

  5. 在下面的单元中输入您的项目 ID。然后运行该单元,确保 Cloud SDK 将适当的项目用于此笔记本中的所有命令。

注意:Jupyter 将前面带 ! 的代码行作为 shell 命令运行,并将前面带 $ 的 Python 变量插入到这些命令中。

PROJECT_ID = "[project-id]" #@param {type:"string"}
! gcloud config set project $PROJECT_ID

验证您的 Google Cloud 帐号

如果您使用的是 AI Platform Notebooks,则您的环境已通过身份验证。请跳过这些步骤。

import sys

# If you are running this notebook in Colab, run this cell and follow the
# instructions to authenticate your Google Cloud account. This provides access
# to your Cloud Storage bucket and lets you submit training jobs and prediction
# requests.

if 'google.colab' in sys.modules:
    from google.colab import auth as google_auth
    google_auth.authenticate_user()

# If you are running this tutorial in a notebook locally, replace the string
# below with the path to your service account key and run this cell to
# authenticate your Google Cloud account.
else:
    %env GOOGLE_APPLICATION_CREDENTIALS your_path_to_credentials.json

# Log in to your account on Google Cloud
!gcloud auth login

导入库

import json
import time
import datetime
from googleapiclient import errors

教程

设置

本部分定义了用于调用 AI Platform Optimizer API 的一些参数和实用程序方法。首先,请填写以下信息。

# Update to your username
USER = '[user-id]' #@param {type: 'string'}

# These will be automatically filled in.
STUDY_ID = '{}_study_{}'.format(USER, datetime.datetime.now().strftime('%Y%m%d_%H%M%S')) #@param {type: 'string'}
REGION = 'us-central1'
def study_parent():
  return 'projects/{}/locations/{}'.format(PROJECT_ID, REGION)

def study_name(study_id):
  return 'projects/{}/locations/{}/studies/{}'.format(PROJECT_ID, REGION, study_id)

def trial_parent(study_id):
  return study_name(study_id)

def trial_name(study_id, trial_id):
  return 'projects/{}/locations/{}/studies/{}/trials/{}'.format(PROJECT_ID, REGION,
                                                                study_id, trial_id)

def operation_name(operation_id):
  return 'projects/{}/locations/{}/operations/{}'.format(PROJECT_ID, REGION, operation_id)

print('USER: {}'.format(USER))
print('PROJECT_ID: {}'.format(PROJECT_ID))
print('REGION: {}'.format(REGION))
print('STUDY_ID: {}'.format(STUDY_ID))

构建 API 客户端

以下单元使用 Google API 发现服务构建自动生成的 API 客户端。JSON 格式的 API 架构托管在 Cloud Storage 存储分区中。

from google.cloud import storage
from googleapiclient import discovery

_OPTIMIZER_API_DOCUMENT_BUCKET = 'caip-optimizer-public'
_OPTIMIZER_API_DOCUMENT_FILE = 'api/ml_public_google_rest_v1.json'

def read_api_document():
  client = storage.Client(PROJECT_ID)
  bucket = client.get_bucket(_OPTIMIZER_API_DOCUMENT_BUCKET)
  blob = bucket.get_blob(_OPTIMIZER_API_DOCUMENT_FILE)
  return blob.download_as_string()

ml = discovery.build_from_document(service=read_api_document())
print('Successfully built the client.')

创建研究配置

下面是一个作为分层 python 字典构建的研究配置示例。它已经填好了。请运行该单元以配置研究。

# Parameter Configuration
param_r = {
    'parameter': 'r',
    'type' : 'DOUBLE',
    'double_value_spec' : {
        'min_value' : 0,
        'max_value' : 1
    }
}

param_theta = {
    'parameter': 'theta',
    'type' : 'DOUBLE',
    'double_value_spec' : {
        'min_value' : 0,
        'max_value' : 1.57
    }
}

# Objective Metrics
metric_y1 = {
    'metric' : 'y1',
    'goal' : 'MINIMIZE'
}

metric_y2 = {
    'metric' : 'y2',
    'goal' : 'MAXIMIZE'
}

# Put it all together in a study configuration
study_config = {
    'algorithm' : 'ALGORITHM_UNSPECIFIED',  # Let the service choose the `default` algorithm.
    'parameters' : [param_r, param_theta,],
    'metrics' : [metric_y1, metric_y2,],
}

study = {'study_config': study_config}
print(json.dumps(study, indent=2, sort_keys=True))

创建研究

接下来,创建研究,您稍后将运行该研究以优化这两个目标。

# Creates a study
req = ml.projects().locations().studies().create(
    parent=study_parent(), studyId=STUDY_ID, body=study)
try :
  print(req.execute())
except errors.HttpError as e:
  if e.resp.status == 409:
    print('Study already existed.')
  else:
    raise e

指标评估函数

接下来,定义一些函数以评估这两个目标指标。

import math

# r * sin(theta)
def Metric1Evaluation(r, theta):
  """Evaluate the first metric on the trial."""
  return r * math.sin(theta)

# r * cose(theta)
def Metric2Evaluation(r, theta):
  """Evaluate the second metric on the trial."""
  return r * math.cos(theta)

def CreateMeasurement(trial_id, r, theta):
  print(("=========== Start Trial: [{0}] =============").format(trial_id))

  # Evaluate both objective metrics for this trial
  y1 = Metric1Evaluation(r, theta)
  y2 = Metric2Evaluation(r, theta)
  print('[r = {0}, theta = {1}] => y1 = r*sin(theta) = {2}, y2 = r*cos(theta) = {3}'.format(r, theta, y1, y2))
  metric1 = {'metric': 'y1', 'value': y1}
  metric2 = {'metric': 'y2', 'value': y2}

  # Return the results for this trial
  measurement = {'step_count': 1, 'metrics': [metric1, metric2,]}
  return measurement

设置运行试验的配置参数

client_id - 请求建议的客户的标识符。如果多个 SuggestTrialsRequests 具有相同的 client_id,则服务将返回相同的建议试验(如果该试验 PENDING),并在建议的最后一个试验已完成的情况下提供一个新试验。

suggestion_count_per_request - 每个请求中请求的建议数(试验数)。

max_trial_id_to_stop - 停止前要探索的试验次数。将其设置为 4 可以缩短运行代码的时间,因此预计不会收敛。如需达到收敛状态,试验次数可能需要大约 20 次(一个好的做法是用总维度乘以 10)。

client_id = 'client1' #@param {type: 'string'}
suggestion_count_per_request =  5 #@param {type: 'integer'}
max_trial_id_to_stop =  50 #@param {type: 'integer'}

print('client_id: {}'.format(client_id))
print('suggestion_count_per_request: {}'.format(suggestion_count_per_request))
print('max_trial_id_to_stop: {}'.format(max_trial_id_to_stop))

运行 AI Platform Optimizer 试验

运行试验。

trial_id = 0
while trial_id < max_trial_id_to_stop:
  # Requests trials.
  resp = ml.projects().locations().studies().trials().suggest(
    parent=trial_parent(STUDY_ID),
    body={'client_id': client_id, 'suggestion_count': suggestion_count_per_request}).execute()
  op_id = resp['name'].split('/')[-1]

  # Polls the suggestion long-running operations.
  get_op = ml.projects().locations().operations().get(name=operation_name(op_id))
  while True:
      operation = get_op.execute()
      if 'done' in operation and operation['done']:
        break
      time.sleep(1)

  for suggested_trial in get_op.execute()['response']['trials']:
    trial_id = int(suggested_trial['name'].split('/')[-1])
    # Featches the suggested trials.
    trial = ml.projects().locations().studies().trials().get(name=trial_name(STUDY_ID, trial_id)).execute()
    if trial['state'] in ['COMPLETED', 'INFEASIBLE']:
      continue

    # Parses the suggested parameters.
    params = {}
    for param in trial['parameters']:
      if param['parameter'] == 'r':
        r = param['floatValue']
      elif param['parameter'] == 'theta':
        theta = param['floatValue']

    # Evaluates trials and reports measurement.
    ml.projects().locations().studies().trials().addMeasurement(
        name=trial_name(STUDY_ID, trial_id),
        body={'measurement': CreateMeasurement(trial_id, r, theta)}).execute()
    # Completes the trial.
    ml.projects().locations().studies().trials().complete(
        name=trial_name(STUDY_ID, trial_id)).execute()

[实验] 直观呈现结果

本部分提供了用于直观呈现上述研究试验的模块。

max_trials_to_annotate = 20

import matplotlib.pyplot as plt
trial_ids = []
y1 = []
y2 = []
resp = ml.projects().locations().studies().trials().list(parent=trial_parent(STUDY_ID)).execute()
for trial in resp['trials']:
  if 'finalMeasurement' in trial:
    trial_ids.append(int(trial['name'].split('/')[-1]))
    metrics = trial['finalMeasurement']['metrics']
    try:
        y1.append([m for m in metrics if m['metric'] == "y1"][0]['value'])
        y2.append([m for m in metrics if m['metric'] == "y2"][0]['value'])
    except:
        pass

fig, ax = plt.subplots()
ax.scatter(y1, y2)
plt.xlabel("y1=r*sin(theta)")
plt.ylabel("y2=r*cos(theta)");
for i, trial_id in enumerate(trial_ids):
  # Only annotates the last `max_trials_to_annotate` trials
  if i > len(trial_ids) - max_trials_to_annotate:
    try:
        ax.annotate(trial_id, (y1[i], y2[i]))
    except:
        pass
plt.gcf().set_size_inches((16, 16))

清理

如需清理此项目中使用的所有 Google Cloud 资源,您可以删除用于本教程的 Google Cloud 项目