Teradata Package for R Function Reference | 17.20 - td_onnx_predict - Teradata Package for R - Look here for syntax, methods and examples for the functions included in the Teradata Package for R.

Teradata® Package for R Function Reference

Deployment
VantageCloud
VantageCore
Edition
Enterprise
IntelliFlex
VMware
Product
Teradata Package for R
Release Number
17.20
Published
March 2024
Language
English (United States)
Last Update
2024-05-03
dita:id
TeradataR_FxRef_Enterprise_1720
Product Category
Teradata Vantage

ONNXPredict

Description

The td_onnx_predict_sqle() function is used to score data in Vantage with a model that has been created outside Vantage and exported to Vantage using ONNX format.

For classical machine learning models on structured data, Vantage has a large set of transformation functions in both the Vantage Analytics Library and Analytics Database Analytic functions. User can use these functions to prepare the input data that the classical machine learning models expect. However, there are no transformation or conversion functions in Vantage to prepare tensors for unstructured data (text, images, video and audio) for ONNX models. The data must be preprocessed before loading to Vantage to conform the tensors into a shape that the ONNX models expect.
As long as the data is in the form expected by your ONNX model, it can be scored by td_onnx_predict_sqle().
td_onnx_predict_sqle() supports models in ONNX format. Several training frameworks support native export functionality to ONNX, such as Chainer, Caffee2, and PyTorch.
User can also convert models from several toolkits like scikit-learn, TensorFlow, Keras, XGBoost, H2O, and Spark ML to ONNX.

Usage

  td_onnx_predict_sqle (
      newdata = NULL,
      modeldata = NULL,
      accumulate = NULL,
      model.output.fields = NULL,
      overwrite.cached.models = "false",
      show.model.input.fields.map = FALSE,
      model.input.fields.map = NULL,
      is.debug = FALSE,
      ...
  )

Arguments

newdata

Required Argument.
Specifies the tbl_teradata containing the input test data.
Types: tbl_teradata

modeldata

Required Argument.
Specifies the tbl_teradata containing the model data to be used for scoring.
Types: tbl_teradata

accumulate

Required Argument.
Specifies the name(s) of input tbl_teradata column(s) to copy to the output.
Types: character OR vector of Strings (character)

model.output.fields

Optional Argument.
Specifies the column(s) of the json output that the user wants to specify as individual columns instead of the entire json_report.
Types: character OR vector of Strings (character)

overwrite.cached.models

Optional Argument.
Specifies the model name that needs to be removed from the cache. When a model loaded into the memory of the node fits in the cache, it stays in the cache until being evicted to make space for another model that needs to be loaded. Therefore, a model can remain in the cache even after the completion of function execution. Other functions that use the same model can use it, saving the cost of reloading it into memory. User should overwrite a cached model only when it is updated, to make sure that the Predict function uses the updated model instead of the cached model.
Note:
Do not use the "overwrite.cached.models" argument except when user is trying to replace a previously cached model. Using the argument in other cases, including in concurrent queries or multiple times within a short period of time lead to an OOM error.
Default Value: "false"
Permitted Values: "true", "t", "yes", "y", "1", "false", "f", "no", "n", "0", "*", "current_cached_model"
Types: character

show.model.input.fields.map

Optional Argument.
Specifies When set to TRUE, the function does not predict the "newdata", instead shows the currently defined fields fully expanded.
Example:
If "model.input.fields.map" is x=[1:4] and "show.model.input.fields.map" is set to TRUE, the output is returned as a varchar column:
model_input_fields_map='x=sepal_len,sepal_wid,petal_len,petal_wid' When "model.input.fields.map" is not specified, the expected default mapping is based on the ONNX inputs defined in the model.
When set to FALSE, which is the default value the function predict the "newdata".
Default Value: FALSE
Types: logical

model.input.fields.map

Optional Argument.
Specifies the output fields to add as individual columns instead of the entire JSON output.
Types: character OR vector of Strings (character)

is.debug

Optional Argument.
Specifies whether debug statements are added to a trace table or not.
When set to TRUE, debug statements are added to a trace table that must be created beforehand.
Notes:

  • Only available with BYOM version 3.00.00.02 and later.

  • To save logs for debugging, user can create an error log by using the is.debug=TRUE parameter in the predict functions.
    A database trace table is used to collect this information which does impact performance of the function, so using small data input sizes is recommended.

  • To generate this log, user must do the following: 1. Create a global trace table with columns vproc_ID BYTE(2), Sequence INTEGER, Trace_Output VARCHAR(31000) 2. Turn on session function tracing: SET SESSION FUNCTION TRACE USING ” FOR TABLE <trace_table_name_created_in_step_1>; 3. Execute function with "is.debug" set to TRUE. 4. Debug information is logged to the table created in step 1. 5. To turn off the logging, either disconnect from the session or run following SQL: SET SESSION FUNCTION TRACE OFF; The trace table is temporary and the information is deleted if user logs off from the session. If long term persistence is necessary, user can copy the table to a permanent table before leaving the session.

Default Value: FALSE
Types: logical

...

Specifies the generic keyword arguments SQLE functions accept. Below are the generic keyword arguments:

persist:
Optional Argument.
Specifies whether to persist the results of the function in a table or not. When set to TRUE, results are persisted in a table; otherwise, results are garbage collected at the end of the session.
Default Value: FALSE
Types: logical

volatile:
Optional Argument.
Specifies whether to put the results of the function in a volatile table or not. When set to TRUE, results are stored in a volatile table, otherwise not.
Default Value: FALSE
Types: logical

Function allows the user to partition, hash, order or local order the input data. These generic arguments are available for each argument that accepts tbl_teradata as input and can be accessed as:

  • "<input.data.arg.name>.partition.column" accepts character or vector of character (Strings)

  • "<input.data.arg.name>.hash.column" accepts character or vector of character (Strings)

  • "<input.data.arg.name>.order.column" accepts character or vector of character (Strings)

  • "local.order.<input.data.arg.name>" accepts logical

Note:
These generic arguments are supported by tdplyr if the underlying SQL Engine function supports, else an exception is raised.

Value

Function returns an object of class "td_onnx_predict_sqle" which is a named list containing object of class "tbl_teradata".
Named list member(s) can be referenced directly with the "$" operator using the name(s):result

Examples

  
    
    # Get the current context/connection..
    con <- td_get_context()$connection
    
    # Load the example data.
    loadExampleData("pmmlpredict_example", "iris_test")
    
    # Create tbl_teradata object.
    iris_test <- tbl(con, "iris_test")
    
    # Set install location of BYOM functions.
    options(byom.install.location = "mldb")
    
    # Check the list of available analytic functions.
    display_analytic_functions(type="BYOM")

    # Create following table on Vantage if it does not exist.
    crt_tbl <- "CREATE SET TABLE byom_models(model_id VARCHAR(40), model BLOB)
                PRIMARY INDEX (model_id);"
    DBI::dbExecute(con, sql(crt_tbl))

    # Run the following query through BTEQ or Teradata Studio to load the
    # models. 'load_byom_model.txt' and byom files can be found under
    # 'inst/scripts' in tdplyr installation directory. This file and the byom
    # models to be loaded should be in the same directory.

    # .import vartext file load_byom_model.txt
    # .repeat *
    # USING (c1 VARCHAR(40), c2 BLOB AS DEFERRED BY NAME) INSERT INTO byom_models(:c1, :c2);

    # Retrieve ONNX model.
    # The 'iris_db_dt_model_sklearn' created with each input variable mapped
    # to a single input tensor, then converted this model into ONNX format
    # with scikit-learn-onnx, and then used to predict the flower species.
    # This model trained using iris_test dataset with scikit-learn.
    skl_model <- tbl(con, "byom_models") 
                 filter(model_id=='iris_db_dt_model_sklearn')

    # The 'iris_db_dt_model_sklearn_floattensor' created by using an input array of
    # four float32 values and named float_input, then converted this model into ONNX
    # format with scikit-learn-onnx, and then used to predict the flower species.
    # This model trained using iris_test dataset with scikit-learn.
    skl_floattensor_model <- tbl(con, "byom_models") 
                             filter(model_id=='iris_db_dt_model_sklearn_floattensor')
    
    # Example 1: Example performs prediction with td_onnx_predict_sqle function
    #            using trained 'skl_model' model in onnx format generated 
    #            outside of Vantage.
    ONNXPredict_out <- td_onnx_predict_sqle(accumulate="id",
                                            newdata=iris_test,
                                            modeldata=skl_model)
    
    # Print the results.
    print(ONNXPredict_out$result)

    # Example 2: Example performs prediction with td_onnx_predict_sqle function using trained
    #            'skl_floattensor_model' model in onnx format generated
    #            outside of Vantage, where input tbl_teradata columns match the order
    #            used when generating the model, by specifying "model.input.fields.map"
    #            to define the columns.
    ONNXPredict_out1 <- td_onnx_predict_sqle(
                         accumulate="id",
                         model.output.fields="output_probability",
                         overwrite.cached.models="*",
                         model.input.fields.map='float_input=sepal_length,
                                                 sepal_width, petal_length,
                                                 petal_width',
                         newdata=iris_test,
                         modeldata=skl_floattensor_model)

    # Print the result.
    print(ONNXPredict_out1$result)
    
    # Example 3: Example to show case the trace table usage using
    #            is.debug=TRUE.
    
    # Create the trace table.
    crt_tbl_query <- 'CREATE GLOBAL TEMPORARY TRACE TABLE BYOM_Trace \
                    (vproc_ID	BYTE(2) \
                    ,Sequence	INTEGER \
                    ,Trace_Output VARCHAR(31000)
                    CHARACTER SET LATIN NOT CASESPECIFIC) \
                    ON COMMIT PRESERVE ROWS;'
    DBI::dbExecute(con, sql(crt_tbl_query))
    
    # Turn on tracing for the session.
    dbExecute(con, "SET SESSION FUNCTION TRACE USING '' FOR TABLE BYOM_Trace;")
    
    # Execute the td_onnx_predict_sqle() function using is.debug=TRUE.
    ONNXPredict_out2 <- td_onnx_predict_sqle(
                                  accumulate="id",
                                  newdata=iris_test,
                                  modeldata=skl_model,
                                  is.debug=TRUE)
    
    # Print the results.
    print(ONNXPredict_out2$result)
    
    # View the trace table information.
    trace_df <- dbGetQuery(con, "select * from BYOM_Trace")
    print(trace_df)
    
    # Turn off tracing for the session.
    dbExecute(con, "SET SESSION FUNCTION TRACE OFF;")