Skip to content

On-premise users: click in-app to access the full platform documentation for your version of DataRobot.

Troubleshooting

The following lists some common issues and how to resolve them.

A job is stuck in INITIALIZING

If using local file intake, make sure you have made a PUT request with the scoring data for the job after the initial POST request.

DataRobot only processes one job at a time per prediction instance, so your job may be queued behind other jobs. Check the job log for details:

curl -X GET https://app.datarobot.com/api/v2/batchPredictions/:id/ \
    -H 'Authorization: Bearer <YOUR_KEY>'

A job is stuck in RUNNING

The job may be running slowly, either because of a slow model or because the scoring data contains errors that the API is trying to identify. You can follow the progress of a job by requesting the job status:

curl -X GET https://app.datarobot.com/api/v2/batchPredictions/:id/ \
    -H 'Authorization: Bearer <YOUR_KEY>'

A job was ABORTED

When a job is aborted, DataRobot logs the reason to the job status. You can check job status from an individual job URL:

curl -X GET https://app.datarobot.com/api/v2/batchPredictions/:id/ \
    -H 'Authorization: Bearer <YOUR_KEY>'

Or from the listing view of all jobs:

curl -X GET https://app.datarobot.com/api/v2/batchPredictions/ \
    -H 'Authorization: Bearer <YOUR_KEY>'

HTTP 406 was returned when uploading a CSV file for local file intake

You are missing the Content-Type: text/csv header.

HTTP 422 was returned when uploading a CSV file for local file intake

You either:

  • Already pushed CSV data for this job. To submit new data, create a new job.
  • Tried to push CSV data for a job that does not require you to push data (e.g., S3 intake).
  • Didn't encode your CSV data in the UTF-8 character set and didn't specify a custom encoding in csvSettings.
  • Didn't encode your CSV data in the proper CSV format and didn't specify a custom format in csvSettings.
  • Tried to push an empty file.

In any of the above cases, the response and the job log will contain an explanation.

Intake stream error due to date format mismatch in Oracle JDBC scoring data

Oracle's DATE type contains a time component, which can cause issues with scoring time series data.

A model trained using the date format yyyy-mm-dd can result in an error for Oracle JDBC scoring data due to Oracle's DATE format.

When DataRobot reads dates from Oracle, the dates are returned in the format yyyy-mm-dd hh:mm:ss by default. This can cause an error when passed to a model expecting a different format.

Use one of the following workarounds to avoid this issue:

  • Train the model using Oracle as the data source to ensure that the time format is the same when scored from Oracle.
  • Use the query option instead of table and schema to allow for the use of SQL functions. Oracle's TO_CHAR function can be used to parse time columns before the data is scored.

The network connection broke while uploading a dataset for local file intake

Create a new job and re-upload the dataset. Failed uploads cannot be resumed and will eventually time out.

The network connection became unavailable while downloading the scoring data for local file output

Re-download the job again. The scored data is available for 48 hours on the managed AI Platform (SaaS) and for 48 hours (but configurable) on the Self-Managed AI Platform (VPC or on-prem).

HTTP 404 was returned while trying to download scored data

You either:

  • Tried to download the scored data for a job that does not have scored data available for download (e.g., S3 output).
  • Started the download before the job had started scoring. In that case, wait until the download link becomes available in the job links and try again.

HTTP 406 was returned when trying to download scored data

Your client sent an Accept header that did not include text/csv. Either do not send the Accept header or include text/csv in it.

CREATE_TABLE scoring fails due to unsupported output column name formats

You may be using a target database as your output adapter that does not support the way DataRobot generates the output format column names. Column names such as name (actual)_PREDICTION when scoring Time Series models might not be supported with all databases.

To work around this issue, you can utilize the Column Name Remapping functionality to re-write the output column name to some form your target database supports.

For instance, if you want to remove the spaces from a column name, you can make a request adding columnNamesRemapping as such:

{
   "deploymentId":"<id>",
   "passthroughColumnsSet":"all",
   "includePredictionStatus":true,
   "intakeSettings":{
      "type":"localFile"
   },
   "outputSettings":{
      "type":"jdbc",
      "dataStoreId":"<id>",
      "credentialId":"<id>",
      "table":"table_name_of_database",
      "schema":"dbo",
      "catalog":"test",
      "statementType":"create_table"
   },
   "columnNamesRemapping":{
      "name (actual)_PREDICTION":"name_actual_PREDICTION"
   }
}

Possible causes for HTTP 422 on job creation

These are the possible causes for an HTTP 422 reply when creating a new Batch Prediction job:

  • You sent an unknown job parameter
  • You specified a job parameter with an unexpected type or value
  • You specified an unknown credential ID in either your intake or output settings
  • You are attempting to score from/to the same S3/Azure/GCP URL (not supported)
  • You are attempting to ingest data from the AI Catalog, but your account does not have access to the AI Catalog
  • You are attempting to ingest data from the AI Catalog and the AI Catalog dataset is not snapshotted (required for predictions) or has not been successfully ingested
  • You are attempting to use a time series custom model (not currently supported)
  • You are attempting to use a traditional time series (ARIMA) model (not currently supported)
  • You requested Prediction Explanations for a multiclass or time series project (not currently supported)
  • You requested prediction warnings for a project other than a regression project (not currently supported)
  • You requested prediction warnings for a project that is not properly configured with prediction boundaries

Updated February 24, 2023