本頁面說明如何使用 AML AI 產生的平台記錄，做為 Cloud Logging 的一部分。AML AI 會使用 Logging API 服務名稱 financialservices.googleapis.com 記錄下列活動：
- 建立引擎設定 (調整)
- 建立模型 (訓練)
- 回測作業
- 預測作業
事前準備
如要查看及管理記錄，請確認您具備正確的身分與存取權管理權限和角色。
平台記錄啟用狀態
反洗錢 AI 的平台記錄一律處於啟用狀態 (無法關閉)。
記錄嚴重性
AML AI 記錄項目會使用三種嚴重性層級：
NOTICE：在作業開始或成功時傳送的項目ERROR是關於失敗作業結束的項目INFO用於作業進度項目
查看平台記錄檔
如要查看平台記錄，請按照下列指示操作：
控制台
如要在 Google Cloud 控制台中查看平台記錄，請按照下列步驟操作：
前往「Logs Explorer」(記錄檔探索工具)：
選取適當的 Google Cloud 專案。
在「Query」欄位中輸入下列查詢指令：
logName=("projects/PROJECT_ID/logs/financialservices.googleapis.com%2Fbacktest" OR "projects/PROJECT_ID/logs/financialservices.googleapis.com%2Fengine_config_creation" OR "projects/PROJECT_ID/logs/financialservices.googleapis.com%2Fmodel_creation" OR "projects/PROJECT_ID/logs/financialservices.googleapis.com%2Fprediction")其中：
PROJECT_ID是您要偵錯或監控的專案 ID。例如：my-project。點選「執行查詢」。
gcloud
gcloud 指令列工具提供 Cloud Logging 的指令列介面。
如要查看專案的記錄，請執行下列指令：
gcloud logging read 'logName=("projects/PROJECT_ID/logs/financialservices.googleapis.com%2Fbacktest" OR
"projects/PROJECT_ID/logs/financialservices.googleapis.com%2Fengine_config_creation" OR
"projects/PROJECT_ID/logs/financialservices.googleapis.com%2Fmodel_creation" OR
"projects/PROJECT_ID/logs/financialservices.googleapis.com%2Fprediction")' --project=PROJECT_ID
其中 PROJECT_ID 是 Google Cloud 專案的 ID。
如要進一步瞭解如何搭配使用 gcloud 工具和 Cloud Logging，請參閱 gcloud logging。
瞭解平台記錄檔
本節說明如何瞭解 AML AI 的特定平台記錄。
開始記錄
作業開始執行時，系統會產生含有 eventKind=START 的記錄。
以下是開始預測執行作業的記錄範例。
jsonPayload:
'@type': type.googleapis.com/google.cloud.financialservices.logging.v1.PredictionLog
engineVersion: projects/PROJECT_ID/locations/REGION_ID/instances/INSTANCE_ID/engineVersions/ENGINE_VERSION_ID
eventKind: START
predictionResult:
dataset: projects/PROJECT_ID/locations/REGION_ID/instances/INSTANCE_ID/datasets/DATASET_ID
endTime: '2023-05-31T00:00:00Z'
model: projects/PROJECT_ID/locations/REGION_ID/instances/INSTANCE_ID/models/MODEL_ID
outputs:
explainabilityDestination:
tableUri: bq://PROJECT_ID.DATASET_ID.EXPLAINABILITY_TABLE_ID
writeDisposition: WRITE_EMPTY
predictionDestination:
tableUri: bq://PROJECT_ID.DATASET_ID.PREDICTION_TABLE_ID
writeDisposition: WRITE_EMPTY
logName: projects/PROJECT_ID/logs/financialservices.googleapis.com%2Fprediction
operation:
first: true
id: projects/PROJECT_NUMBER/locations/REGION_ID/operations/OPERATION_ID
producer: financialservices.googleapis.com
receiveTimestamp: '2023-06-07T12:30:48.417285528Z'
resource:
labels:
instance_id: INSTANCE_ID
location: REGION_ID
prediction_result_id: PREDICTION_ID
resource_container: projects/PROJECT_NUMBER
type: financialservices.googleapis.com/PredictionResult
您可以在記錄檔探索工具的「查詢」欄位中新增其他指令，進一步縮小顯示的記錄範圍。
新增下列指令，即可在所選資料集上顯示所有已啟動的預測執行作業：
logName="projects/PROJECT_ID/logs/financialservices.googleapis.com%2Fprediction" AND jsonPayload.predictionResult.dataset="projects/PROJECT_ID/locations/REGION_ID/instances/INSTANCE_ID/datasets/DATASET_ID" AND jsonPayload.eventKind="START"
進度記錄
含有 eventKind=PROGRESS 的記錄會提供作業進度資訊。
以下是建立模型的記錄範例。completedTaskCount 與 taskCount 可用於估算模型訓練的進度。
jsonPayload:
'@type': type.googleapis.com/google.cloud.financialservices.logging.v1.ModelCreationLog
completedTaskCount: 11
engineVersion: projects/PROJECT_ID/locations/REGION_ID/instances/INSTANCE_ID/engineVersions/ENGINE_VERSION_ID
eventKind: PROGRESS
model:
endTime: '2023-05-31T00:00:00Z'
engineConfig: projects/PROJECT_ID/locations/REGION_ID/instances/INSTANCE_ID/engineConfigs/ENGINE_CONFIG_ID
engineVersion: projects/PROJECT_ID/locations/REGION_ID/instances/INSTANCE_ID/engineVersions/ENGINE_VERSION_ID
lineOfBusiness: RETAIL
primaryDataset: projects/PROJECT_ID/locations/REGION_ID/instances/INSTANCE_ID/datasets/DATASET_ID
state: CREATING
partyCount: '9246'
taskCount: 16
logName: projects/PROJECT_ID/logs/financialservices.googleapis.com%2Fmodel_creation
operation:
id: projects/PROJECT_NUMBER/locations/REGION_ID/operations/OPERATION_ID
producer: financialservices.googleapis.com
receiveTimestamp: '2023-06-07T13:57:00.454668648Z'
resource:
labels:
instance_id: INSTANCE_ID
location: REGION_ID
model_id: MODEL_ID
resource_container: projects/PROJECT_NUMBER
type: financialservices.googleapis.com/Model
severity: INFO
timestamp: '2023-06-07T13:56:59.772973055Z'
結束記錄
作業結束時，系統會產生含有 eventKind=END 的記錄。
以下是引擎設定建立失敗的記錄範例。這份報告包含提供資料集中錯誤資料的錯誤。
jsonPayload:
'@type': type.googleapis.com/google.cloud.financialservices.logging.v1.EngineConfigCreationLog
completedTaskCount: 3
engineConfig:
engineVersion: projects/PROJECT_ID/locations/REGION_ID/instances/INSTANCE_ID/engineVersions/ENGINE_VERSION_ID
lineOfBusiness: RETAIL
performanceTarget:
partyInvestigationsPerPeriodHint: '100'
state: CREATING
tuning:
endTime: '2019-04-30T00:00:00Z'
primaryDataset: projects/PROJECT_ID/locations/REGION_ID/instances/INSTANCE_ID/datasets/DATASET_ID
eventKind: END
operationStatus:
code: 9
details:
- '@type': type.googleapis.com/google.rpc.ErrorInfo
domain: financialservices.googleapis.com
metadata:
count: '15'
data_field: party_id, validity_start_time
data_table: party
description: There is a duplicate primary key value in the database resulting
in unique key violation. Note that for tables with validity_start_time,
the primary key includes validity_start_time
test: GROUP BY party_id, validity_start_time HAVING count(1) > 1
reason: DUPLICATE_PRIMARY_KEY
message: Dataset validation failed with 1 error. See error details for individual
violations.
partyCount: '9246'
taskCount: 16
logName: projects/PROJECT_ID/logs/financialservices.googleapis.com%2Fengine_config_creation
operation:
id: projects/PROJECT_NUMBER/locations/REGION_ID/operations/OPERATION_ID
last: true
producer: financialservices.googleapis.com
receiveTimestamp: '2023-06-07T14:26:30.214382295Z'
resource:
labels:
engine_config_id: ENGINE_CONFIG_ID
instance_id: INSTANCE_ID
location: REGION_ID
resource_container: projects/PROJECT_NUMBER
type: financialservices.googleapis.com/EngineConfig
severity: ERROR
timestamp: '2023-06-07T14:26:29.670913895Z'
如要查看所有建立引擎設定錯誤記錄，請使用下列篩選器：
logName="projects/PROJECT_ID/logs/financialservices.googleapis.com%2Fengine_config_creation" AND severity>=ERROR