Davis AI service
The Davis® predictive and causal AI platform service delivers a comprehensive suite of AI/ML and core statistical capabilities seamlessly integrated into the Dynatrace platform. These essential predictive and causal analysis functions empower application creators to effectively address various use cases, including time series forecasting, learning anomaly detection models for time series data, and automated monitoring of metric behavior changes and anomalies.
By leveraging the Davis® AI capabilities, app developers gain direct and streamlined access to the entire AI/ML functionalities. This direct integration allows seamless utilization of these AI/ML capabilities within their respective apps, workflows, or Dynatrace functions, optimizing the efficiency and effectiveness of their development processes.
Concepts
You need to know the following concepts to work with the Davis® AI service.
Analyzer definitions
Every analyzer has a static definition, which describes its capabilities. The definition consists of the following parts: general metadata, input, and output definition.
The general metadata section contains essential information about the analyzer, like its name, displayName, and description. The name of an analyzer is fully qualified and guaranteed not to change in the future.
Input and output definitions share a standard schema describing how the input and output of a specific analyzer are structured. You can retrieve the exact structure of the input and output objects for a particular analyzer from the analyzers/<Analyzer-Name>/json-schema/input or the analyzers/<Analyzer-Name>/json-schema/output endpoints.
Analyzer execution
Flow
To execute an analyzer for the first time, we recommend the following request flow:
-
Query all analyzers: You should query all the available analyzers and choose among them using the endpoint.
-
Query a specific analyzer: Query a particular definition of an analyzer to see which functional scope and parameters the selected analyzer supports using the endpoint.
-
Query schema: Query the input and result schema of the analyzer using the and endpoint. This will provide you with a JSON schema for the analyzer you're interested in.
-
Execute the Davis analyzer: Use the endpoint to execute a Davis analyzer. Davis analyzers are executed asynchronously. The analyzers, however, return a synchronous result if the defined wait period isn't exceeded.
Execution logs
An analyzer provides execution logs if the execution fails or if there is more auxiliary information about the result. Before an analyzer is executed, preliminary checks, such as input validations, are done. If these checks fail, the analyzer isn't executed, and an error is returned.
If the input matches the input definition, the request will always produce a consumable, successful result.
It can still happen, however, that some analysis inside the analyzer fails due to other reasons. In such cases, the analyzer provides execution logs pointing toward the problem.
Execution logs have a specific level, indicating the severity of the issue for the given analyzer. These levels can be filtered when executing an analyzer.
Example
The dt.statistics.GenericForecastAnalyzer produces a forecast for a given time series and is an excellent tool for predicting business metrics.
Given the following input:
{
"timeSeriesData": "timeseries avg(dt.host.cpu.usage)"
}
This analyzer returns the forecast predictions for the given time series:
{
"result": {
"resultId": "b433d5b1e3125ff3",
"resultStatus": "SUCCESSFUL",
"executionStatus": "COMPLETED",
"input": {
"generalParameters": {
"timeframe": "unknown-unknown",
"logVerbosity": "WARNING",
"resolveDimensionalQueryData": false
},
"forecastHorizon": 10,
"timeSeriesData": {
"expression": "timeseries avg(dt.host.cpu.usage)"
},
"coverageProbability": 0.9,
"nPaths": 200
},
"output": [
{
"timeSeriesDataWithPredictions": {
"records": [
{
"dt.davis.forecast:lower": [
20.302533297777046, 20.30119956974749, 20.299826093391346, 20.298413637147934, 20.296962988044072,
20.295474948510375, 20.293950333306114, 20.29238996657088, 20.290794679018166, 20.289165305282822
],
"dt.davis.forecast:upper": [
20.531565739653274, 20.5316575385106, 20.53178908569177, 20.531959612757507, 20.532168332681056,
20.53241444303185, 20.532697129050682, 20.53301556659801, 20.533368924960413, 20.53375636950308
],
"dt.davis.forecast:point": [
20.4170495226752, 20.416428558113733, 20.41580759355227, 20.415186628990803, 20.414565664429336,
20.41394469986787, 20.413323735306403, 20.412702770744936, 20.412081806183473, 20.411460841622006
],
"interval": "60000000000",
"timeframe": {
"end": "2023-04-03T05:42Z",
"start": "2023-04-03T05:32Z"
},
"dt.davis.internal.dataName": "avg(dt.host.cpu.usage)"
}
],
"types": [
{
"indexRange": [0, 0],
"mappings": {
"dt.davis.forecast:lower": {
"type": "array",
"types": [
{
"indexRange": [0, 9],
"mappings": {
"element": {
"type": "double"
}
}
}
]
},
"timeframe": {
"type": "timeframe"
},
"dt.davis.forecast:upper": {
"type": "array",
"types": [
{
"indexRange": [0, 9],
"mappings": {
"element": {
"type": "double"
}
}
}
]
},
"dt.davis.forecast:point": {
"type": "array",
"types": [
{
"indexRange": [0, 9],
"mappings": {
"element": {
"type": "double"
}
}
}
]
},
"interval": {
"type": "duration"
},
"dt.davis.internal.dataName": {
"type": "string"
}
}
}
]
},
"forecastQualityAssessment": "valid",
"analysisStatus": "OK",
"analyzedTimeSeriesQuery": {
"expression": "timeseries avg(dt.host.cpu.usage)",
"timeframe": {
"startTime": "2023-04-03T03:31:32.718Z",
"endTime": "2023-04-03T05:31:32.718Z"
}
}
}
]
}
}