Skip to content

Click in-app to access the full platform documentation for your version of DataRobot.

Batch prediction use cases

The following provides several end-to-end examples of scoring with API code for both CSV files and external services.

Note

These use cases require the DataRobot API client to be installed.

End-to-end scoring of CSV files from local files

The following example scores a local CSV file, waits for processing to start, and then initializes the download.

import datarobot as dr

dr.Client(
    endpoint="https://app.datarobot.com/api/v2",
    token="...",
)

deployment_id = "..."

input_file = "to_predict.csv"
output_file = "predicted.csv"

job = dr.BatchPredictionJob.score_to_file(
    deployment_id,
    input_file,
    output_file,
    passthrough_columns_set="all"
)

print("started scoring...", job)
job.wait_for_completion()

Prediction Explanations

You can include Prediction Explanations by adding the desired Prediction Explanation parameters to the job configuration:

job = dr.BatchPredictionJob.score_to_file(
    deployment_id,
    input_file,
    output_file,
    max_explanations=10,
    threshold_high=0.5,
    threshold_low=0.15,
)

Custom CSV format

If your CSV files does not match the default CSV format, you can modify the expected CSV format by setting csvSettings:

job = dr.BatchPredictionJob.score_to_file(
    deployment_id,
    input_file,
    output_file,
    csv_settings={
        'delimiter': ';',
        'quotechar': '\'',
        'encoding': 'ms_kanji',
    },
)

End-to-end scoring of CSV files on S3

import datarobot as dr

dr.Client(
    endpoint="https://app.datarobot.com/api/v2",
    token="...",
)

deployment_id = "616d01a8ddbd17fc2c75caf4"
credential_id = "..."

s3_csv_input_file = 's3://my-bucket/data/to_predict.csv'
s3_csv_output_file = 's3://my-bucket/data/predicted.csv'

job = dr.BatchPredictionJob.score_s3(
    deployment_id,
    source_url=s3_csv_input_file,
    destination_url=s3_csv_output_file,
    credential=credential_id
)

print("started scoring...", job)
job.wait_for_completion()

The same functionality is available for score_azure and score_gcp. You can also specify the credential object itself, instead of a credential ID:

credentials = dr.Credential.get(credential_id)

job = dr.BatchPredictionJob.score_s3(
    deployment_id,
    source_url=s3_csv_input_file,
    destination_url=s3_csv_output_file,
    credential=credentials,
)

Prediction Explanations

You can include Prediction Explanations by adding the desired Prediction Explanation parameters to the job configuration:

job = dr.BatchPredictionJob.score_s3(
    deployment_id,
    source_url=s3_csv_input_file,
    destination_url=s3_csv_output_file,
    credential=credential_id,
    max_explanations=10,
    threshold_high=0.5,
    threshold_low=0.15,
)

AI Catalog-to-CSV file scoring

When using the AI Catalog for intake, you need the dataset_id of an already created dataset.

import datarobot as dr

dr.Client(
    endpoint="https://app.datarobot.com/api/v2",
    token="...",
)

deployment_id = "616d01a8ddbd17fc2c75caf4"
credential_id = "..."
dataset_id = "..."

dataset = dr.Dataset.get(dataset_id)

job = dr.BatchPredictionJob.score(
    deployment_id,
    intake_settings={
        'type': 'dataset',
        'dataset_id': dataset,
    },
    output_settings={
        'type': 'localFile',
    },
)

job.wait_for_completion()

End-to-end scoring from a JDBC PostgreSQL database

The following reads a scoring dataset from the table public.scoring_data and saves the scored data back to public.scored_data (assuming that table already exists).

import datarobot as dr

dr.Client(
    endpoint="https://app.datarobot.com/api/v2",
    token="...",
)

deployment_id = "616d01a8ddbd17fc2c75caf4"
credential_id = "..."
datastore_id = "..."

intake_settings = {
    'type': 'jdbc',
    'table': 'scoring_data',
    'schema': 'public',
    'data_store_id': datastore_id,
    'credential_id': credential_id,
}

output_settings = {
    'type': 'jdbc',
    'table': 'scored_data',
    'schema': 'public',
    'data_store_id': datastore_id,
    'credential_id': credential_id,
    'statement_type': 'insert'
}

job = dr.BatchPredictionJob.score(
    deployment_id,
    passthrough_columns_set='all',
    intake_settings=intake_settings,
    output_settings=output_settings,
)

print("started scoring...", job)
job.wait_for_completion()

More details about JDBC scoring can be found here.

End-to-end scoring with Snowflake

The following example reads a scoring dataset from the table public.SCORING_DATA and saves the scored data back to public.SCORED_DATA (assuming that table already exists).

import datarobot as dr
dr.Client(
    endpoint="https://app.datarobot.com/api/v2",
    token="...",
)
deployment_id = "616d01a8ddbd17fc2c75caf4"
credential_id = "..."
cloud_storage_credential_id = "..."
datastore_id = "..."
intake_settings = {
    'type': 'snowflake',
    'table': 'SCORING_DATA',
    'schema': 'PUBLIC',
    'external_stage': 'my_s3_stage_in_snowflake',
    'data_store_id': datastore_id,
    'credential_id': credential_id,
    'cloud_storage_type': 's3',
    'cloud_storage_credential_id': cloud_storage_credential_id
}
output_settings = {
    'type': 'snowflake',
    'table': 'SCORED_DATA',
    'schema': 'PUBLIC',
    'statement_type': 'insert'
    'external_stage': 'my_s3_stage_in_snowflake',
    'data_store_id': datastore_id,
    'credential_id': credential_id,
    'cloud_storage_type': 's3',
    'cloud_storage_credential_id': cloud_storage_credential_id
}
job = dr.BatchPredictionJob.score(
    deployment_id,
    passthrough_columns_set='all',
    intake_settings=intake_settings,
    output_settings=output_settings,
)
print("started scoring...", job)
job.wait_for_completion()

More details about Snowflake scoring can be found in intake and output documentation.

End-to-end scoring with Synapse

The following example reads a scoring dataset from the table public.scoring_data and saves the scored data back to public.scored_data (assuming that table already exists).

import datarobot as dr
dr.Client(
    endpoint="https://app.datarobot.com/api/v2",
    token="...",
)
deployment_id = "616d01a8ddbd17fc2c75caf4"
credential_id = "..."
cloud_storage_credential_id = "..."
datastore_id = "..."
intake_settings = {
    'type': 'synapse',
    'table': 'SCORING_DATA',
    'schema': 'PUBLIC',
    'external_data_source': 'some_datastore',
    'data_store_id': datastore_id,
    'credential_id': credential_id,
    'cloud_storage_credential_id': cloud_storage_credential_id
}
output_settings = {
    'type': 'synapse',
    'table': 'SCORED_DATA',
    'schema': 'PUBLIC',
    'statement_type': 'insert'
    'external_data_source': 'some_datastore',
    'data_store_id': datastore_id,
    'credential_id': credential_id,
    'cloud_storage_credential_id': cloud_storage_credential_id
}
job = dr.BatchPredictionJob.score(
    deployment_id,
    passthrough_columns_set='all',
    intake_settings=intake_settings,
    output_settings=output_settings,
)
print("started scoring...", job)
job.wait_for_completion()

More details about Synapse scoring can be found in the intake and output documentation.

End-to-end scoring with BigQuery

The following example scores data from a BigQuery table and sends results to a BigQuery table.

import datarobot as dr

dr.Client(
    endpoint="https://app.datarobot.com/api/v2",
    token="...",
)

deployment_id = "616d01a8ddbd17fc2c75caf4"
gcs_credential_id = "6166c01ee91fb6641ecd28bd"

intake_settings = {
    'type': 'bigquery',
    'dataset': 'my-dataset',
    'table': 'intake-table',
    'bucket': 'my-bucket',
    'credential_id': gcs_credential_id,
}

output_settings = {
    'type': 'bigquery',
    'dataset': 'my-dataset',
    'table': 'output-table',
    'bucket': 'my-bucket',
    'credential_id': gcs_credential_id,
}

job = dr.BatchPredictionJob.score(
    deployment=deployment_id,
    intake_settings=intake_settings,
    output_settings=output_settings,
    include_prediction_status=True,
    passthrough_columns=["some_col_name"],
)

print("started scoring...", job)
job.wait_for_completion()

More details about BigQuery scoring can be found in the intake and output documentation.


Updated November 30, 2021
Back to top