Search⌘ K
AI Features
Home
Courses
Mastering Artificial Intelligence with Azure Cognitive Services

Multivariate Anomaly Detection - Testing and Best Practices

Explore how to implement and test multivariate anomaly detection using Azure Decision Cognitive Services. Understand model training, asynchronous detection processes, deleting models, and best practices for preparing and handling time-series data effectively.

We'll cover the following...

To recap, in the previous lesson, we prepared and uploaded the dataset to Azure blob storage and generated the SAS URL. Then, we submitted a request for model training to detect the multi-variate anomalies in the dataset. After that, we have checked the status of the model.

Detecting anomalies using the trained model

Once the model is trained successfully, it’s ready to detect the anomalies in the dataset.

C++
blob_data_sas_url = "<YOUR_BLOB_SAS_URL_FOR_ZIP_FILE>"
start_time = datetime(2021, 1, 1, 0, 0, 0)
end_time = datetime(2021, 1, 2, 12, 0, 0)
detection_req = DetectionRequest(source = blob_data_sas_url, start_time = start_time, end_time = end_time)
response_header = anomaly_detector_client.detect_anomaly(multi_variate_trained_model_id, 
                                                         detection_req,
                                                        cls = lambda *args: [args[i] for i in range(len(args))]
                                                    )[-1]
result_id = response_header['Location'].split("/")[-1]
results = anomaly_detector_client.get_detection_result(result_id)
while results.summary.status != DetectionStatus.READY and results.summary.status != DetectionStatus.FAILED:
    results = anomaly_detector_client.get_detection_result(result_id)
    time.sleep(1)
if results.summary.status == DetectionStatus.FAILED:
    if results.summary.errors:
        for error in results.summary.errors:
            print("The Error code: {}. ".format(error.code))
            print("The Error Message: {}".format(error.message))
anomaly_detect_results = []
for res in results.results:
    anomaly = {}
    if res.value is not None:
        value = res.value
        is_anomaly = value.is_anomaly
        if is_anomaly is True:
            anomaly["timestamp"] = res.timestamp.strftime("%m/%d/%Y, %H:%M:%S")
            anomaly["variable_contributing"] = []
            anomaly["severity"] = value.severity
            for contributing_factor in value.contributors:
                anomaly["variable_contributing"].append(contributing_factor.variable)
            anomaly_detect_results.append(anomaly)            
print("Total number of anomalies identified: ", len(anomaly_detect_results))
print("Anomaly Identified:")
for anomaly in anomaly_detect_results:
    print(anomaly)

Explanation:

  • Lines 1–3, are from the earlier code. You will need to paste the blob SAS URL in line 1.

  • In line 5, we create the object DetectionRequest class and pass the data source URL and the time-frames within which we want our model to detect the anomalies.

  • From lines 7–10, we ...