Skip to main content

Tips and Tricks

Data preparation​

The data submitted to Azure Anomaly Detector is expected to be time series data. It must be an exact time series. There are a couple of techniques to prepare your data for Anomaly Detector.

Database data bucketing​

Most database support some sort of data bucketing, the following is an example of data bucketing with Azure SQL. The following query will bucket the data into 10 minute buckets.

DECLARE @period INT = 10

SELECT
    CONVERT(VARCHAR(20), CONVERT(DATETIMEOFFSET, DATE_BUCKET(minute, @period, timestamp)), 127) AS timestamp,
    MAX([temperature]) as temperature,
    MAX([humidity]) as humidity,
    MAX([prediction]) as prediction
FROM telemetry
WHERE
    deviceId = @deviceId AND timestamp > DATEADD(day, -@numberOfDays, GETDATE())
GROUP BY
    deviceId, DATE_BUCKET(minute, @period, timestamp);

Pandas data resampling​

Pandas is a popular Python library for data manipulation and analysis. Refer to the Pandas documentation for more information on resampling.

import pandas as pd
import numpy as np

rng = pd.date_range("1/1/2012", periods=100, freq="S")
ts = pd.Series(np.random.randint(0, 500, len(rng)), index=rng)
rs = ts.resample("5Min").max()