def afd_train_model_demo()

in demo_scripts/create_afd_resources.py [0:0]

• 92 lines of code
• 9 McCabe index (conditional complexity)

def afd_train_model_demo(s3bucket, s3suffix, eventtype, modelname, iamrole, fraudcategory):
    
    #############################################
    #####               Setup               #####
    config_file = RECIPE[fraudcategory]
                
    MODEL_NAME = modelname
    EVENT_TYPE = eventtype
    IAM_ROLE = iamrole
    
    EVENT_VARIABLES = [variable["variable_name"] for variable in config_file["variable_mappings"]]
    EVENT_LABELS = ["fraud", "legit"]

    # Variable mappings of demo data in this use case.  Important to teach this to customer
    click.echo(f'{pd.DataFrame(config_file["variable_mappings"])}')
    
    # Copy file from 
    data_path = config_file["data_path"]
    response = s3.list_objects_v2(Bucket=s3bucket, Prefix=os.path.join(s3suffix, data_path))
    if response['KeyCount'] == 0:
        s3.put_object(
            Bucket=s3bucket, 
            Key=os.path.join(s3suffix, data_path),
            Body=open(data_path, 'rb')
        )
    S3_DATA_PATH = "s3://" + os.path.join(s3bucket, s3suffix, data_path)
       
    #############################################
    ##### Create event variables and labels #####
    
    # -- create variable  --
    for variable in config_file["variable_mappings"]:
        
        DEFAULT_VALUE = '0.0' if variable["data_type"] == "FLOAT" else '<null>'
        
        try:
            resp = client.get_variables(name = variable["variable_name"])
            click.echo("{0} exists, data type: {1}".format(variable["variable_name"], resp['variables'][0]['dataType']))
        except:
            click.echo("Creating variable: {0}".format(variable["variable_name"]))
            resp = client.create_variable(
                    name         = variable["variable_name"],
                    dataType     = variable["data_type"],
                    dataSource   ='EVENT',
                    defaultValue = DEFAULT_VALUE, 
                    description  = variable["variable_name"],
                    variableType = variable["variable_type"])
            
    response = client.put_label(
        name = "fraud",
        description = "FRAUD")
    
    response = client.put_label(
        name = "legit",
        description = "LEGIT")
        

    #############################################
    #####   Define Entity and Event Types   #####
    
    # -- create entity type --
    try:
        response = client.get_entity_types(name = ENTITY_TYPE)
        click.echo("-- entity type exists --")
        click.echo(response)
    except:
        response = client.put_entity_type(
            name        = ENTITY_TYPE,
            description = ENTITY_DESC
        )
        click.echo("-- create entity type --")
        click.echo(response)


    # -- create event type --
    try:
        response = client.get_event_types(name = EVENT_TYPE)
        click.echo("\n-- event type exists --")
        click.echo(response)
    except:
        response = client.put_event_type (
            name           = EVENT_TYPE,
            eventVariables = EVENT_VARIABLES,
            labels         = EVENT_LABELS,
            entityTypes    = [ENTITY_TYPE])
        click.echo("\n-- create event type --")
        click.echo(response)


    #############################################
    #####   Create and train your model     #####
    try:
        response = client.create_model(
           description   = MODEL_DESC,
           eventTypeName = EVENT_TYPE,
           modelId       = MODEL_NAME,
           modelType     = MODEL_TYPE)
        click.echo("-- initalize model --")
        click.echo(response)
    except Exception:
        pass
    
    # -- initalized the model, it's now ready to train --
    
    # -- first define training_data_schema for model to use --
    training_data_schema = {
        'modelVariables' : EVENT_VARIABLES,
        'labelSchema'    : {
            'labelMapper' : {
                'FRAUD' : ["fraud"],
                'LEGIT' : ["legit"]
            }
        }
    }
    
    response = client.create_model_version(
        modelId             = MODEL_NAME,
        modelType           = MODEL_TYPE,
        trainingDataSource  = 'EXTERNAL_EVENTS',
        trainingDataSchema  = training_data_schema,
        externalEventsDetail = {
            'dataLocation'     : S3_DATA_PATH,
            'dataAccessRoleArn': IAM_ROLE
        }
    )
    model_version = response['modelVersionNumber']
    click.echo("-- model training --")
    click.echo(response)