com.datarobot.mlops_spark_utils

Class MLOpsSparkUtils

java.lang.Object

com.datarobot.mlops_spark_utils.MLOpsSparkUtils

public class MLOpsSparkUtils
extends Object

Provide utility functions to perform MLOps monitoring in Spark.

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	MLOpsSparkUtils.PredictionsDataParameters
static class	MLOpsSparkUtils.PredictionStatisticsParameters
static class	MLOpsSparkUtils.RawPredictionDataParameters

Constructor Summary

Constructors

Constructor and Description
MLOpsSparkUtils()

Method Summary

Modifier and Type	Method and Description
static void	reportActuals(org.apache.spark.sql.Dataset<org.apache.spark.sql.Row> dataFrame, String deploymentId, String modelId, String channelConfig) Report actual values for predictions.
static void	reportPredictions(org.apache.spark.sql.Dataset<org.apache.spark.sql.Row> dataFrame, String deploymentID, String modelId, String channelConfig, Double scoringTime, List<String> targetColumnNames) Report predictions and features given a DataFrame (Dataset of Row).
static void	reportPredictions(org.apache.spark.sql.Dataset<org.apache.spark.sql.Row> dataFrame, String deploymentID, String modelId, String channelConfig, Double scoringTime, List<String> targetColumnNames, Integer predictionsServiced) Report predictions and features given a DataFrame (Dataset of Row).
static void	reportPredictions(org.apache.spark.sql.Dataset<org.apache.spark.sql.Row> dataFrame, String deploymentID, String modelId, String channelConfig, Double scoringTime, List<String> targetColumnNames, String assocIdColName) Report predictions and features given a DataFrame (Dataset of Row).
static void	reportPredictions(org.apache.spark.sql.Dataset<org.apache.spark.sql.Row> dataFrame, String deploymentID, String modelId, String channelConfig, Double scoringTime, List<String> targetColumnNames, String assocIdColName, Integer predictionsServiced) Report predictions and features given a DataFrame (Dataset of Row).
static void	reportPredictions(org.apache.spark.sql.Dataset<org.apache.spark.sql.Row> dataFrame, String deploymentID, String modelId, String channelConfig, Double scoringTime, String[] targetColumnNames) Report predictions and features given a DataFrame (Dataset of Row).
static void	reportPredictions(org.apache.spark.sql.Dataset<org.apache.spark.sql.Row> dataFrame, String deploymentID, String modelId, String channelConfig, Double scoringTime, String[] targetColumnNames, Integer predictionsServiced) Report predictions and features given a DataFrame (Dataset of Row).
static void	reportPredictions(org.apache.spark.sql.Dataset<org.apache.spark.sql.Row> dataFrame, String deploymentID, String modelId, String channelConfig, Double scoringTime, String[] targetColumnNames, String assocIdColName) Report predictions and features given a DataFrame (Dataset of Row).
static void	reportPredictions(org.apache.spark.sql.Dataset<org.apache.spark.sql.Row> dataFrame, String deploymentID, String modelId, String channelConfig, Double scoringTime, String[] targetColumnNames, String assocIdColName, Integer predictionsServiced) Report predictions and features given a DataFrame (Dataset of Row).

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

MLOpsSparkUtils

public MLOpsSparkUtils()

Method Detail

reportPredictions

public static void reportPredictions(org.apache.spark.sql.Dataset<org.apache.spark.sql.Row> dataFrame,
                                     String deploymentID,
                                     String modelId,
                                     String channelConfig,
                                     Double scoringTime,
                                     String[] targetColumnNames,
                                     String assocIdColName,
                                     Integer predictionsServiced)
                              throws com.datarobot.mlops.common.exceptions.DRCommonException

Report predictions and features given a DataFrame (Dataset of Row). This method performs all the necessary steps to report the predictions and features provided in the DataFrame parameter to DataRobot MLOps. The parameters provided are used to tell this helper function how to connect to MLOps and the structure of the DataFrame.

This variant of the function gets the target columns as a String[] and accepts association ids and allows the caller to override the number of predictions serviced, which by default is the number of predictions included in the dataframe.

Parameters:

dataFrame - The DataFrame to report

deploymentID - ID of the deployment this data belongs to in MLOps DataRobot

modelId - Model Id used for predictions

channelConfig - A string containing the MLOps channel configuration

scoringTime - The time it took to perform scoring in milliseconds.

targetColumnNames - An array of column names to use for the target of the prediction. If only one item is provided, this is assumed to be a regression prediction. If more than one, it is a classification prediction.

assocIdColName - The name of the column containing the association ids for the predictions

predictionsServiced - The number of predictions actually serviced. This may be higher than the number of predictions reported here in the dataframe if downsampling is applied before reporting.

Throws:

com.datarobot.mlops.common.exceptions.DRCommonException

reportPredictions

public static void reportPredictions(org.apache.spark.sql.Dataset<org.apache.spark.sql.Row> dataFrame,
                                     String deploymentID,
                                     String modelId,
                                     String channelConfig,
                                     Double scoringTime,
                                     String[] targetColumnNames,
                                     String assocIdColName)
                              throws com.datarobot.mlops.common.exceptions.DRCommonException

Report predictions and features given a DataFrame (Dataset of Row). This method performs all the necessary steps to report the predictions and features provided in the DataFrame parameter to DataRobot MLOps. The parameters provided are used to tell this helper function how to connect to MLOps and the structure of the DataFrame.

This variant of the function gets the target columns as a String[] and accepts association ids.

Parameters:

dataFrame - The DataFrame to report

deploymentID - ID of the deployment this data belongs to in MLOps DataRobot

modelId - Model Id used for predictions

channelConfig - A string containing the MLOps channel configuration

scoringTime - The time it took to perform scoring in milliseconds.

targetColumnNames - An array of column names to use for the target of the prediction. If only one item is provided, this is assumed to be a regression prediction. If more than one, it is a classification prediction.

assocIdColName - The name of the column containing the association ids for the predictions

Throws:

com.datarobot.mlops.common.exceptions.DRCommonException

reportPredictions

public static void reportPredictions(org.apache.spark.sql.Dataset<org.apache.spark.sql.Row> dataFrame,
                                     String deploymentID,
                                     String modelId,
                                     String channelConfig,
                                     Double scoringTime,
                                     List<String> targetColumnNames,
                                     String assocIdColName,
                                     Integer predictionsServiced)
                              throws com.datarobot.mlops.common.exceptions.DRCommonException

Report predictions and features given a DataFrame (Dataset of Row). This method performs all the necessary steps to report the predictions and features provided in the DataFrame parameter to DataRobot MLOps. The parameters provided are used to tell this helper function how to connect to MLOps and the structure of the DataFrame.

This variant of the function gets the target columns as a List of String and allows the caller to override the number of predictions serviced, which by default is the number of predictions included in the dataframe.

Parameters:

dataFrame - The DataFrame to report

deploymentID - Id of the deployment this data belongs to in DataRobot MLOps

modelId - Model Id used for predictions

channelConfig - A string containing MLOps channel configuration

scoringTime - Time it took to perform scoring in milliseconds

targetColumnNames - An array of column names to use for the target of the prediction. If only one item is provided, this is assumed to be a regression prediction. If more than one, it is a classification prediction.

assocIdColName - The name of the column containing the association ids for the predictions

predictionsServiced - The number of predictions actually serviced. This may be higher than the number of predictions reported here in the dataframe if downsampling is applied before reporting.

Throws:

com.datarobot.mlops.common.exceptions.DRCommonException

reportPredictions

public static void reportPredictions(org.apache.spark.sql.Dataset<org.apache.spark.sql.Row> dataFrame,
                                     String deploymentID,
                                     String modelId,
                                     String channelConfig,
                                     Double scoringTime,
                                     List<String> targetColumnNames,
                                     String assocIdColName)
                              throws com.datarobot.mlops.common.exceptions.DRCommonException

Report predictions and features given a DataFrame (Dataset of Row). This method performs all the necessary steps to report the predictions and features provided in the DataFrame parameter to DataRobot MLOps. The parameters provided are used to tell this helper function how to connect to MLOps and the structure of the DataFrame.

This variant of the function gets the target columns as a List of String.

Parameters:

dataFrame - The DataFrame to report

deploymentID - Id of the deployment this data belongs to in DataRobot MLOps

modelId - Model Id used for predictions

channelConfig - A string containing MLOps channel configuration

scoringTime - Time it took to perform scoring in milliseconds

targetColumnNames - An array of column names to use for the target of the prediction. If only one item is provided, this is assumed to be a regression prediction. If more than one, it is a classification prediction.

assocIdColName - The name of the column containing the association ids for the predictions

Throws:

com.datarobot.mlops.common.exceptions.DRCommonException

reportPredictions

public static void reportPredictions(org.apache.spark.sql.Dataset<org.apache.spark.sql.Row> dataFrame,
                                     String deploymentID,
                                     String modelId,
                                     String channelConfig,
                                     Double scoringTime,
                                     String[] targetColumnNames,
                                     Integer predictionsServiced)
                              throws com.datarobot.mlops.common.exceptions.DRCommonException

Report predictions and features given a DataFrame (Dataset of Row). This method performs all the necessary steps to report the predictions and features provided in the DataFrame parameter to DataRobot MLOps. The parameters provided are used to tell this helper function how to connect to MLOps and the structure of the DataFrame.

This variant of the function gets the target columns as a String[] and does not accept association ids and allows the caller to override the number of predictions serviced, which by default is the number of predictions included in the dataframe.

Parameters:

dataFrame - The DataFrame to report

deploymentID - Id of the deployment this data belongs to in DataRobot MLOps

modelId - Model Id used for predictions

channelConfig - A string containing MLOps channel configuration

scoringTime - Time it took to perform scoring in milliseconds

targetColumnNames - An array of column names to use for the target of the prediction. If only one item if provided, this is assumed to be a regression prediction. If more than one, it is a classification prediction.

predictionsServiced - The number of predictions actually serviced. This may be higher than the number of predictions reported here in the dataframe if downsampling is applied before reporting.

Throws:

com.datarobot.mlops.common.exceptions.DRCommonException

reportPredictions

public static void reportPredictions(org.apache.spark.sql.Dataset<org.apache.spark.sql.Row> dataFrame,
                                     String deploymentID,
                                     String modelId,
                                     String channelConfig,
                                     Double scoringTime,
                                     String[] targetColumnNames)
                              throws com.datarobot.mlops.common.exceptions.DRCommonException

Report predictions and features given a DataFrame (Dataset of Row). This method performs all the necessary steps to report the predictions and features provided in the DataFrame parameter to DataRobot MLOps. The parameters provided are used to tell this helper function how to connect to MLOps and the structure of the DataFrame.

This variant of the function gets the target columns as a String[] and does not accept association ids.

Parameters:

dataFrame - The DataFrame to report

deploymentID - Id of the deployment this data belongs to in DataRobot MLOps

modelId - Model Id used for predictions

channelConfig - A string containing MLOps channel configuration

scoringTime - Time it took to perform scoring in milliseconds

targetColumnNames - An array of column names to use for the target of the prediction. If only one item if provided, this is assumed to be a regression prediction. If more than one, it is a classification prediction.

Throws:

com.datarobot.mlops.common.exceptions.DRCommonException

reportPredictions

public static void reportPredictions(org.apache.spark.sql.Dataset<org.apache.spark.sql.Row> dataFrame,
                                     String deploymentID,
                                     String modelId,
                                     String channelConfig,
                                     Double scoringTime,
                                     List<String> targetColumnNames,
                                     Integer predictionsServiced)
                              throws com.datarobot.mlops.common.exceptions.DRCommonException

Report predictions and features given a DataFrame (Dataset of Row). This method performs all the necessary steps to report the predictions and features provided in the DataFrame parameter to DataRobot MLOps. The parameters provided are used to tell this helper function how to connect to MLOps and the structure of the DataFrame.

This variant of the function gets the target columns as a List of String, does not accept association ids, and allows the caller to override the number of predictions serviced, which by default is the number of predictions included in the dataframe.

Parameters:

dataFrame - The DataFrame to report

deploymentID - Id of the deployment this data belongs to in DataRobot MLOps

modelId - Model Id used for predictions

channelConfig - A string containing MLOps channel configuration

scoringTime - Time it took to perform scoring in milliseconds

targetColumnNames - An array of column names to use for the target of the prediction. If only one item if provided, this is assumed to be a regression prediction. If more then one, it is a classification prediction.

predictionsServiced - The number of predictions actually serviced. This may be higher than the number of predictions reported here in the dataframe if downsampling is applied before reporting.

Throws:

com.datarobot.mlops.common.exceptions.DRCommonException

reportPredictions

public static void reportPredictions(org.apache.spark.sql.Dataset<org.apache.spark.sql.Row> dataFrame,
                                     String deploymentID,
                                     String modelId,
                                     String channelConfig,
                                     Double scoringTime,
                                     List<String> targetColumnNames)
                              throws com.datarobot.mlops.common.exceptions.DRCommonException

Report predictions and features given a DataFrame (Dataset of Row). This method performs all the necessary steps to report the predictions and features provided in the DataFrame parameter to DataRobot MLOps. The parameters provided are used to tell this helper function how to connect to MLOps and the structure of the DataFrame.

This variant of the function gets the target columns as a List of String and does not accept association ids.

Parameters:

dataFrame - The DataFrame to report

deploymentID - Id of the deployment this data belongs to in DataRobot MLOps

modelId - Model Id used for predictions

channelConfig - A string containing MLOps channel configuration

scoringTime - Time it took to perform scoring in milliseconds

targetColumnNames - An array of column names to use for the target of the prediction. If only one item if provided, this is assumed to be a regression prediction. If more then one, it is a classification prediction.

Throws:

com.datarobot.mlops.common.exceptions.DRCommonException

reportActuals

public static void reportActuals(org.apache.spark.sql.Dataset<org.apache.spark.sql.Row> dataFrame,
                                 String deploymentId,
                                 String modelId,
                                 String channelConfig)
                          throws com.datarobot.mlops.common.exceptions.DRCommonException

Report actual values for predictions. The DataFrame provided should contain all the needed information about the predictions and the actual result.
The following columns are expected to exist in the DataFrame (order does not matter):

• "associationId": (String) a unique identifier used with a predicted row
• "actualValue": the actual value of a prediction. This should be numeric for regression models and a string label for classification models.
• "wasActedOn": (Boolean)(Optional) Indicates if the prediction was acted on in a way that could have affected the actual outcome. For example, if a hospital patient is predicted to be readmitted in 30 days, extra procedures or new medication might be given to mitigate this problem, influencing the actual outcome of the prediction.
• "timestamp": the datetime when actual values were obtained, formatted according to RFC3339.

Parameters:

dataFrame - the DataFrame containing the actuals as specified above

deploymentId - The deployment id these actuals belong to

modelId - The model id that used for the predictions

channelConfig - MLOps channel configuration string

Throws:

com.datarobot.mlops.common.exceptions.DRCommonException