Application Configuration ========================= ML Monitoring Application can be set up and customized by authoring a JSON configuration. The configuration then needs to be saved in an object store location and passed in the CONFIG_FILE variable of RUNTIME_PARAMETER while starting a job run. This document demonstrates how to define the application components and create a application configuration. Sample Config File ---------------------- .. collapse:: ml-monitoring-config.json .. code-block:: json { "monitor_id": "", "storage_details": { "storage_type": "OciObjectStorage", "params": { "namespace": "", "bucket_name": "", "object_prefix": "" } }, "input_schema": { "Age": { "data_type": "integer", "variable_type": "continuous", "column_type": "input" }, "EnvironmentSatisfaction": { "data_type": "integer", "variable_type": "continuous", "column_type": "input" } }, "baseline_reader": { "type": "CSVDaskDataReader", "params": { "file_path": "oci://" } }, "prediction_reader": { "type": "CSVDaskDataReader", "params": { "data_source": { "type": "ObjectStorageFileSearchDataSource", "params": { "file_path": ["oci://"], "filter_arg": [ { "partition_based_date_range": { "start": "2023-06-26", "end": "2023-06-27", "data_format": ".d{4}-d{2}-d{2}." } } ] } } }, "dataset_metrics": [ { "type": "RowCount" } ], "feature_metrics": { "Age": [ { "type": "Min" }, { "type": "Max" } ], "EnvironmentSatisfaction": [ { "type": "Mode" }, { "type": "Count" } ] }, "transformers": [ { "type": "ConditionalFeatureTransformer", "params": { "conditional_features": [ { "feature_name": "Young", "data_type": "integer", "variable_type": "ordinal", "expression": "df.Age < 30" } ] } } ], "post_processors": [ { "type": "SaveMetricOutputAsJsonPostProcessor", "params": { "file_name": "", "test_results_file_name": "", "file_location_expression": "", "date_range": { "start": "2023-08-01", "end": "2023-08-05" }, "can_overwrite_profile_json": false, "can_overwrite_test_results_json": false, "namespace": "", "bucket_name": "" } }, { "type": "OCIMonitoringApplicationPostProcessor", "params": { "compartment_id": "", "namespace": "", "date_range": { "start": "2023-08-01", "end": "2023-08-05" }, "dimensions": { "key1": "value1", "key2": "value2" } } } ], "tags": { "tag": "value" } }, "test_config": { "tags": { "key_1": "these tags are sent in test results" }, "feature_metric_tests": [ { "feature_name": "Age", "tests": [ { "test_name": "TestGreaterThan", "metric_key": "Min", "threshold_value": 17 }, { "test_name": "TestIsComplete" } ] } ], "dataset_metric_tests": [ { "test_name": "TestGreaterThan", "metric_key": "RowCount", "threshold_value": 40, "tags": { "subtype": "falls-xgb" } } ] } } | ML Monitoring Application Components ------------------------------------ Monitor ID ~~~~~~~~~~~~~~~~ **This is a required component and must be defined in the config.** User provided id used to identify a monitor config uniquely. Below are the rules to define a monitor_id. * The length should be minimum 8 characters and maximum 48 characters. * Valid characters are letters (upper or lowercase), numbers, hyphens, underscores, and periods. Description ^^^^^^^^^^^ .. list-table:: :widths: 25 25 50 :header-rows: 1 * - Key - Value - Example * - monitor_id - user defined string - "monitor_id": "speech_model_monitor" Example ^^^^^^^^^^^^^ .. code-block:: json {"monitor_id": "speech_model_monitor"} Storage Details ~~~~~~~~~~~~~~~~ **This is a required component and must be defined in the config.** Details of the type of storage and location for retrieving the baseline profile(in case of a prediction run) and persist the internal state of a run. Description ^^^^^^^^^^^^^ .. list-table:: :widths: 100 100 100 :header-rows: 1 * - Field Name - Description - Example * - storage_type - type of storage to be used for storing the internal state - "storage_type": "OciObjectStorage" * - param - params (required) - "params": { "namespace": "", "bucket_name": "", "object_prefix": "" } Supported Storage Details ^^^^^^^^^^^^^^^^^^^^^^^^^^ * OciObjectStorage Required Parameters * namespace - namespace of the bucket * bucket_name - bucket name Optional Parameters * object_prefix - prefix for creating the directory for saving the internal state of the runs Example ^^^^^^^^^^^^^ .. collapse:: storage_details .. code-block:: json "storage_details": { "storage_type": "OciObjectStorage", "params": { "namespace": "", "bucket_name": "", "object_prefix": "" } } Input Schema ~~~~~~~~~~~~~~~~ **This is a required component and must be defined in the config.** Input schema is the map of features and their data types, variable types, and column type. Description ^^^^^^^^^^^ .. list-table:: :widths: 25 25 50 :header-rows: 1 * - Key - Value - Example * - feature_name - object of key value pair of data_type ,variable type and column_type - "Age": { "data_type": "integer", "variable_type": "continuous", "column_type": "input" } * **Data Type (Required)** Data types can be provided for each feature of the input dataset which represent the type of the feature value. *Supported data_type* - "integer", "float", "string", "boolean", "text", "object" * **Variable Type (Required)** Variable types can be provided for each feature of the input dataset which represent the type of a statistical random variable. *Supported variable_type* - "continuous", "discrete", "nominal", "ordinal", "binary", "text", "object" * **Column Type (Optional - Default value "input")** Insights supports performance metrics for regression and classification models. In addition to these, Insights also supports multivariate metrics like Feature Importance. These metrics require the prediction columns or target columns (ground truth) to be in the input dataset. To make it easier to configure the metrics, Insights allows users to configure the prediction or target columns using the feature schema. *Supported column_type* - "input", "prediction", "target", "prediction_score" Example ^^^^^^^^^^^^^ .. code-block:: json { "input_schema": { "sepal length (cm)": { "data_type": "float", "variable_type": "continuous", "column_type": "input" }, "sepal width (cm)": { "data_type": "float", "variable_type": "continuous" "column_type": "input" } } } BASELINE READER ~~~~~~~~~~~~~~~~~ **If the action type is RUN_BASELINE, this is a required component and must be defined in the config.** The baseline_reader allows for the ingestion of raw data into the framework for a baseline run. Description ^^^^^^^^^^^^^ .. list-table:: :widths: 100 100 100 :header-rows: 1 * - Field Name - Description - Example * - type - type of reader to be used - "type": "JsonlDaskDataReader" * - param - reader params data_source (optional) - "params": { "file_path": "oci://.csv" } "data_source": { "type": "ObjectStorageFileSearchDataSource", "params": { "file_path": [ "oci://@//dataset.csv" ], "filter_arg": [ { "partition_based_date_range": { "start": "2023-06-26", "end": "2023-06-27", "data_format": ".d{4}-d{2}-d{2}." } } ] } } Example using data source for determining the data location ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ .. code-block:: json "baseline_reader": { "type": "CSVDaskDataReader", "params": { "data_source": { "type": "ObjectStorageFileSearchDataSource", "params": { "file_path": [ "oci://@//dataset.csv" ], "filter_arg": [ { "partition_based_date_range": { "start": "2023-06-26", "end": "2023-06-27", "data_format": ".d{4}-d{2}-d{2}." } } ] } } } } Example without using data_source ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ .. code-block:: json { "baseline_reader": { "type": "CSVDaskDataReader", "params": { "file_path": "oci://.csv" } } } **Supported Reader** .. collapse:: Supported Readers .. code-block:: python CSVDaskDataReader JsonlDaskDataReader NestedJsonDaskDataReader ADWApplicationDataReader We can use reader params to define the location of the files to be read or can specify a data source in the reader. **Data Source** The Data Source component is responsible for interacting with a specific data source and returning a list of locations to be read. .. collapse:: Supported Data Sources .. code-block:: python OCIObjectStorageDataSource OCIDatePrefixDataSource ObjectStorageFileSearchDataSource PREDICTION READER ~~~~~~~~~~~~~~~~~ **If the action type is RUN_PREDICTION, this is a required component and must be defined in the config.** The prediction_reader allows for the ingestion of raw data into the framework for a prediction run. Description ^^^^^^^^^^^^^ .. list-table:: :widths: 100 100 100 :header-rows: 1 * - Field Name - Description - Example * - type - type of reader to be used - "type": "JsonlDaskDataReader" * - param - reader params (required) data_source (optional) - "params": { "file_path": "oci://.csv" } "data_source": { "type": "ObjectStorageFileSearchDataSource", "params": { "file_path": [ "oci://@//dataset.csv" ], "filter_arg": [ { "partition_based_date_range": { "start": "2023-06-26", "end": "2023-06-27", "data_format": ".d{4}-d{2}-d{2}." } } ] } } Example using data source for determining the data location ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ .. code-block:: json "prediction_reader": { "type": "CSVDaskDataReader", "params": { "data_source": { "type": "ObjectStorageFileSearchDataSource", "params": { "file_path": [ "oci://@//dataset.csv" ], "filter_arg": [ { "partition_based_date_range": { "start": "2023-06-26", "end": "2023-06-27", "data_format": ".d{4}-d{2}-d{2}." } } ] } } } } Example without using data_source ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ .. code-block:: json { "prediction_reader": { "type": "CSVDaskDataReader", "params": { "file_path": "oci://.csv" } } } **Supported Reader** .. collapse:: Supported Readers .. code-block:: python CSVDaskDataReader JsonlDaskDataReader NestedJsonDaskDataReader ADWApplicationDataReader We can use reader params to define the location of the files to be read or can specify a data source in the reader. **Data Source** The Data Source component is responsible for interacting with a specific data source and returning a list of locations to be read. .. collapse:: Supported Data Sources .. code-block:: python OCIObjectStorageDataSource OCIDatePrefixDataSource ObjectStorageFileSearchDataSource Feature Metrics ~~~~~~~~~~~~~~~~~~ In this section, you need to add metrics that you need for each feature. Description ^^^^^^^^^^^^^ .. list-table:: :widths: 25 25 :header-rows: 1 * - Key - Value * - feature_name - metric list **Supported Feature Metrics** `More Metric details `_ .. collapse:: Supported Feature Metrics .. code-block:: python # Data quality metrics Count DistinctCount DuplicateCount FrequencyDistribution Max Mean Min Mode ProbabilityDistribution Range Skewness StandardDeviation Sum IQR Kurtosis TopKFrequentElements TypeMetric Variance IsPositive IsNegative IsNonZero Percentiles # Data Integrity IsConstantFeature IsQuasiConstantFeature Quartiles # Drift Metrics KullbackLeibler KolmogorovSmirnov ChiSquare JensenShannon PopulationStabilityIndex # Bias and Fairness ClassImbalance # Date Time Metrics DateTimeMin DateTimeMax DateTimeDuration | Example ^^^^^^^^^^^^^ .. code-block:: json "feature_metric": { "sepal length (cm)" : [ {"type": "Sum"},{"type": "Quartiles"} ], "sepal width (cm)": [ {"type": "Min"},{"type": "DistinctCount"} ], "petal length (cm)": [ {"type": "Count"},{"type": "Mean"} ], "petal width (cm)": [ {"type": "IsQuasiConstantFeature"},{"type": "Kurtosis"} ] } Dataset Metrics ~~~~~~~~~~~~~~~~ Description ^^^^^^^^^^^^^ List of metrics to be calculated on the data set. Example ^^^^^^^^ .. code-block:: json "data_set_metric": [ { "type": "RowCount" } ] **Supported Data Set Metrics** `More Metric details `_ .. collapse:: Supported Data Set Metrics .. code-block:: python # Data Quality Metrics CramersVCorrelation PearsonCorrelation CorrelationRatio # Regression Metrics RowCount MeanAbsoluteError MeanSquaredError R2Score RootMeanSquaredError MeanSquaredLogError MeanAbsolutePercentageError MaxError # Classification metrics AccuracyScore PrecisionScore RecallScore FBetaScore FalsePositiveRate FalseNegativeRate Specificity ConfusionMatrix LogLoss ROCCurve ROCAreaUnderCurve PrecisionRecallCurve PrecisionRecallAreaUnderCurve # Conflict Metrics ConflictPrediction ConflictLabel | Post Processor ~~~~~~~~~~~~~~~~~ Post processor components are responsible for running any action after the entire data set is processed and all the metrics are calculated. Description ^^^^^^^^^^^^^ .. list-table:: :widths: 100 100 100 100 :header-rows: 1 * - Field Name - Description - Example1 - Example2 * - type - type of post processor - "type": "SaveMetricOutputAsJsonPostProcessor" - "type": "OCIMonitoringApplicationPostProcessor" * - param - post processor params (required) - "params": { "file_name": "profile.json", "test_results_file_name": "test_result.json", "file_location_expression": "bug-bash/mlm/profile-$start_$end.json", "date_range": { "start": "2023-08-01", "end": "2023-08-05" }, "can_overwrite_profile_json": false, "can_overwrite_test_results_json": false, "namespace": "", "bucket_name": "" } - "params": { "compartment_id": "", "namespace": "", "date_range": { "start": "2023-08-01", "end": "2023-08-05" }, "dimensions": { "key1": "value1", "key2": "value2" } } Example ^^^^^^^^ .. code-block:: json "post_processors": [ { "type": "SaveMetricOutputAsJsonPostProcessor", "params": { "file_name": "profile.json", "test_results_file_name": "test_result.json", "file_location_expression": "bug-bash/mlm/profile-$start_$end.json", "date_range": { "start": "2023-08-01", "end": "2023-08-05" }, "can_overwrite_profile_json": false, "can_overwrite_test_results_json": false, "namespace": "", "bucket_name": "" } } ] Supported Post Processors ^^^^^^^^^^^^^^^^^^^^^^^^^ * ``SaveMetricOutputAsJsonPostProcessor`` This will store the metric result output in user provided Object storage location in a json format. Required Parameters * **bucket_name** - The name of the OCI Object Storage bucket. * **namespace** - The OCI Object Storage namespace. Optional Parameters * file_location_expression - The expression of the object location within the bucket, which would be configured as per the date_range argument. * if file_location_expression is not provided and no date_range is provided in runtime parameter, object location is generated by the application as '/MLM///file_name.json' * if file_location_expression is not provided and date_range, object location is generated by the application as - '/MLM///$start-$end/' * **file_name** - A filename for the object name. Default value for file_name is 'profile.json' * **can_overwrite_profile_json** - A boolean whether the existing profile file should be overwritten. By default the profile file would Not be overwritten. * **test_results_file_name** - A filename for the Test result object name. Default value for file_name is 'test_result.json' * **can_overwrite_test_results_json** - A boolean whether the existing test result file should be overwritten. By default the test result file would Not be overwritten. * **date_range** - A dictionary containing optional date range which would be configured in file location. This can be overwritten by passing START and END DATE in RUNTIME_PARAMETER. **Example** .. code-block:: json "post_processors": [ { "type": "SaveMetricOutputAsJsonPostProcessor", "params": { "file_name": "profile.json", "test_results_file_name": "test_result.json", "file_location_expression": "/usecase/$start_$end", "date_range": { "start": "2023-08-01", "end": "2023-08-05" }, "can_overwrite_profile_json": true, "can_overwrite_test_results_json": false, "namespace": "", "bucket_name": "" } } ] In the above example, the JSON result would be stored at the location - /usecase/2023-08-01_2023-08-05/profile.json and Test Results would be stored at the location - /usecase/2023-08-01_2023-08-05/test_result.json * ``OCIMonitoringApplicationPostProcessor`` This will will push the Ml Insight Test Suite results to OCI Monitoring Service in user provided Compartment Id Required Parameters * **compartment_id** - The OCID of the compartment to use for metrics. Optional Parameters * **dimensions** - Additional dimensions for the metrics (default is an empty). * **namespace** - The namespace for the OCI monitoring (default is 'ml_monitoring'). * **date_range** - A dictionary containing optional date range which would be configured in file location. This can be overwritten by passing START and END DATE in RUNTIME_PARAMETER. **Example** .. code-block:: json "post_processors": [ { "type": "OCIMonitoringApplicationPostProcessor", "params": { "compartment_id": "", "namespace": "", "date_range": { "start": "2023-08-01", "end": "2023-08-05" }, "dimensions": { "key1": "value1", "key2": "value2" } } } ] In the above example, Ml Insight Test Suite results will be pushed to user provided compartment_id * ``SaveMetricOutputAsJsonPostProcessor`` For details, please refer here :doc:`user_guide/adw/adw_writer` Transformer ~~~~~~~~~~~~~ The transformer component provides an easy way to do simple in-memory transformations on the input data. The list of transformers to be used to add a conditional feature or transform the data before insights run. Description ^^^^^^^^^^^^ .. list-table:: :widths: 100 100 100 :header-rows: 1 * - Field Name - Description - Example * - type - type of transformer - "type": "ConditionalFeatureTransformer" * - param - conditional_features - List of conditional features - "params": { "conditional_features": [ { "feature_name": "Young", "data_type": "integer", "variable_type": "ordinal", "expression": "df.Age < 30" } ] } Conditional Features ~~~~~~~~~~~~~~~~~~~~~ .. list-table:: :widths: 100 100 100 :header-rows: 1 * - Field Name - Value - Remarks * - expression - Python expression, to be written using pandas series based functions. Only pandas series level functions are supported in a python expression and the symbol 'df'. - the expression must return a valid output. For example: "expression": "df.Age < 30" * - feature_name - - * - data_type - The data type of the feature. - * - variable_type - The variable type of the feature. - Example ^^^^^^^^ .. code-block:: json "transformers": [ { "type": "ConditionalFeatureTransformer", "params": { "conditional_features": [ { "feature_name": "Young", "data_type": "integer", "variable_type": "continuous", "expression": "int(json_row['Age'] < 30)" } ] } } ] Tags ~~~~~~~ Note, this is a application internal concept and should not be confused with OCI resource tags. * User provided key value pair * Users can provide tags to be associated with a profile. For eg: when running the baseline/prediction run, we can store: <"tenancy": "tenancy-xyz"> Example ^^^^^^^^ .. code-block:: json "tags": { "tenancy": "tenancy-xyz" } Tests Config ~~~~~~~~~~~~ For detailed documentation, please refer to section: `Test Config `_