Azure Blob Storage

Set up an import job to ingest data into Arize from Azure
You will need to contact [email protected] to set up Azure Blob Storage. Refer to step 3 for more details.
Set up an import job to log inference files to Arize. Updates to files are checked every 10 seconds. Users generally find a sweet spot around a few hundred thousand to a million rows in each file, with the total file limit being 1GB.

Step 1. Get The Storage Container Name & Prefix

Create a blob storage container and folder (optional) where you would like Arize to pull your model's inferences.
For example you might set up a container named bucket1 and folder /click-thru-rate/production/v1/ that contains CSV files of your model inferences.
In this example, your bucket name is bucket1 and your prefix is click-thru-rate/production/v1/
There are multiple ways to structure model data. To easily ingest model inference data from storage, adopt a standardized directory structure across all models.

Step 2. Get the Storage Account Name & Shared Key

Find the storage account name that your container is created under, and choose one access key to share with Arize.
In Azure UI, from the Container page, navigate back to the storage account in the top left.
To get a shared key, navigate to Access Keys in the storage account menu.

Step 3. Contact Arize To Complete The Permissions Setup

Email [email protected] to set up Azure permissions.
Our team will contact you to securely share your container name and storage account name, and provide your access key. Once we confirm setup is complete, begin setting up import jobs.

Step 4. Select Azure Storage

Navigate to the 'Upload Data' page on the left navigation bar in the Arize platform. From there, select the 'Azure Blob Storage' card to begin a new file import job.
Fill in the file path where you would like Arize to pull your model's inferences. Arize will automatically infer your bucket name and prefix.
Example File Path In Arize UI
In this example, you might have a bucket and folder named azure://example-demo-bucket/click-thru-rate/production/v1/ that contains parquet files of your model inferences. Your bucket name is example-demo-bucket and your prefix is click-thru-rate/production/v1/.
The file structure can take into consideration various model environments (training, production, etc) and locations of ground truth.
Example 1: Predictions & Actuals Stored in Separate Folders (different prefixes)
This example contains model predictions and actuals in separate files. This helps in cases of delayed actuals.
├── prediction-folder/
│ ├── 12-1-2022.parquet #this file can contain multiple versions
│ ├── 12-2-2022.parquet
│ ├── 12-3-2022.parquet
├── actuals-folder/
│ ├── 12-1-2022.parquet
│ ├── 12-2-2022.parquet
│ └── 12-3-2022.parquet
Example 2: Production & Training Stored in Separate Folders
This example separates model environments (production and training).
├── production-folder/
│ ├── 12-1-2022.parquet
│ ├── 12-2-2022.parquet
│ ├── 12-3-2022.parquet
├── training-folder/
│ ├── 12-1-2022.parquet
│ ├── 12-2-2022.parquet
│ └── 12-3-2022.parquet

Step 5. Add Proof Of Ownership To Your Container

In your container metadata, add an entry with the key as arize_ingestion_key and the provided tag value.
  • In Arize UI: Copy the arize_ingestion_key value.
  • In Azure UI: Navigate to your Container -> Settings -> Metadata.
Click on Metadata and fill out the key value pair defined in the Arize UI

Step 6a. Define Your Model Schema

Model schema parameters are a way of organizing model inference data to ingest to Arize. When configuring your schema, be sure to match your data column headers with the model schema.
You can either use a form or a simple JSON-based schema to specify the column mapping.
Arize supports CSV, Parquet, Avro, and Apache Arrow. Refer here for a list of the expected data types by input type.
The unique identifier of a specific prediction. Limited to 128 characters.
The timestamp of the prediction in seconds or an RFC3339 timestamp
Optional, defaults to current timestamp at file ingestion time
Column name for the prediction value
Required based on model type
Column name for the predicted score
Required based on model type
Column name for the actual or ground truth value
Optional for production records
Column name for the ground truth score
Required based on model type
Column name for ranking groups or lists in ranking models
Required for ranking models
Column name for rank of each element on the its group or list
Required for ranking models
Column name for ranking actual or ground truth value
Required for ranking models
Column name for ranking ground truth score
Required for ranking models
A string prefix to describe a column feature/. Features must be sent in the same file as predictions
Arize automatically infers columns as features. Choose between feature prefixing OR inferred features.
A string prefix to describe a column tag/. Tags must be sent in the same file as predictions and features
A string prefix to describe a column shap/. SHAP must be sent in the same file as predictions or with a matching prediction_id
A column to specify model version. version/ assigns a version to the corresponding data within a column, or configure your version within the UI
Optional, defaults to 'no_version'
Distinguish different batches of data under the same model_id and model_version. Must be specified as a constant during job setup or in the schema
Optional for validation records only
A list of columns to exclude if the features property is not included in the ingestion schema
A list of embedding columns, required vector column, optional raw data column, and optional link to data column. Learn more here

Step 6b. Validate Your Model Schema

Once you fill in your applicable predictions, actuals, and model inputs, click 'Validate Schema' to visualize your model schema in the Arize UI. Check that your column names and corresponding data match for a successful import job.
Once finished, your import job will be created and will start polling your bucket for files.
If your model receives delayed actuals, connect your predictions and actuals using the same prediction ID, which links your data together in the Arize platform. Arize regularly checks your data source for both predictions and actuals, and ingests them separately as they become available. Learn more here.

Step 7. Check Your File Import Job

Arize will attempt a dry run to validate your job for any access, schema or record-level errors. If the dry run is successful, you may then create the import job.
After creating a job following a successful dry run, you will be taken to the 'Job Status' tab where you can see the status of your import jobs. A created job will regularly sync new data from your data source with Arize. You can view the job details and import progress by clicking on the job ID, which uncovers more information about the job.

Step 8. Troubleshoot Import Job

An import job may run into a few problems. Use the dry run and job details UI to troubleshoot and quickly resolve data ingestion issues.

Validation Errors

If there is an error validating a file against the model schema, Arize will surface an actionable error message. From there, click on the 'Fix Schema' button to adjust your model schema.

Dry Run File/Table Passes But The Job Fails

If your dry run is successful, but your job fails, click on the job ID to view the job details. This uncovers job details such as information about the file path or query id, the last import job, potential errors, and error locations.
Once you've identified the job failure point, fix the file errors and reupload the file to Arize with a new name.