options.md

Description

drctrl can operate some datarobot features (project building, training, prediction, freezing model) by specifying configuration file.

Each commands must be specified in top-level field of yaml configuration file.

environment:
   options:

fit:
   options:

frozen:
   oprions:

predict:
   options:

environment

• project setting field
• if project is not supplied, automaticaly create a new project based on given options

basic usage

create a new project

environment:
   project_id: 
   project_name: 'sample project'
   target_feature: target
   metric: AUC
   cv_method: 'random'
   validation_type: 'CV'
   validation_params: 
      holdout_pct: 20
      reps: 3
      seed: 2017
   dataset:
       type: file
       path: ./data/raw/
       filename: boston.csv
   autopilot: 'manual' 

When project is given, other options are not required.

environment:
  project_id: <project_id>

options

• project_id
  • if null is specified, automaticaly create a new project based on given options
  • type: str or none
  • required: False
• project_name
  • type: str
  • required: False
• target_feature
  • target feature name
  • type: str
  • required: True
• metric
  • optimization metric
    • valid metrics depends on project type (classification, regression)
    • see datarobot api metrics
  • type: str
  • required: True
• cv_method
  • partitioning method (it is prefix of datarobot partitioning class name)
    • datetime, user, straitified, random, user, group
    • see data robot api doc
  • type: str
  • required: True
• validation_type
  • partition type
    • CV : cross validation
    • TVH : train, validation holdoutset
  • type: str
  • required: True
• validation_params
  • this option depends on cv_method and validation_type
    • required options can be confirmed at data robot api doc
    • practically, you should try making a copy of partition settings from already exist project by using drctrl get_partition_setting option.
  • type: map
  • required: True
  • allowempty: True
• dataset
  • dataset option
    • see io plugins
  • type: map
  • required: True
  • allowempty: True
• autopilot
  • run autopilot when creating project
    • fullauto
    • manual
  • type: str
  • required: True
• convert_features

  • convert feature type

    • give following format

    cconvert_features:
      - {name: <source_feature_name>, rename_to: <dest_name>, variable_type: <type>}
      - {name: <source_feature_name>, rename_to: <dest_name>, variable_type: <type>}
      - {name: <source_feature_name>, rename_to: <dest_name>, variable_type: <type>}

  • type: sequence

  • required: False

fit

basic usage

fit:
   model_id: <model_id> or none
   source_featurelist: 'Raw Features'

options

• model_id

  • if model_id is null, autopilot is executed.
  • type: str or none
  • required: True

• sample_pct

  • the percentage of the project dataset used in training the model
  • type: float
  • required: False

• autopilot

  • if model_id is not specified, this option is required.
  • type: str
  • required: False

• featurelist_name

  • featurelist name
  • type: str
  • required: False

• source_featurelist

  • type: str
  • required: True

• except_features

  • create featurelist without specified features
  • required: False
  • type: sequence

     except_features:
       - 'feature_2'
       - 'feature_1'

• wait_for_completion

  • wait for model creation or autopilot completion.
  • type: bool
  • required: False

frozen

basic usage

frozen:
   model_id: <model_id>
   sample_pct: 50.0

options

• model_id
  • if None, target model is automaticaly selected based on the order of validation
  • type: str or null
  • required: True
• sample_pct
  • the percentage of the project dataset used in training the model
  • type: float
  • required: False
• training_row_count
  • only required for models in datetime partitioned projects. If specified, defines the number of rows used to train the model and evaluate backtest scores.
  • type: int
  • required: False
• training_duration
  • only required for models in datetime partitioned projects. a duration string specifying the duration spanned by the data used to train the model and evaluate backtest scores.
  • type: int
  • required: False
• start_date
  • only required for models in datetime partitioned projects.
  • format: yyyymmdd_hhmmss
  • type: str
  • required: False
• end_date
  • only required for models in datetime partitioned projects.
  • format: yyyymmdd_hhmmss
  • type: str
  • required: False

predict

basic usage

predict:
   model_id: <model_id> or null
   input:
       type: file
       path: './data/raw/'
       filename: boston.csv
   output:
       type: file
       path: './'
       filename: result.csv

options

• model_id
  • if None, target model is automaticaly selected based on the order of validation
  • type: str or null
  • required: True
• del_dataset
  • delete dataset when prediction is completed
  • type: bool
  • required: False
• reasoncode
  • if True, reasoncode will be calculated with the model.
  • type: bool
  • required: False
• max_codes
  • only required for calculating reasoncode. 1-10 can be specified.
  • type: int
  • required: False
• merge_origin
  • if True, prediction result is merged with given dataset specified in input field.
  • type: bool
  • required: False
• feature_impact
  • if True, feature impact will be calculated.
    • when reasoncode is specified, this option will be specified.
  • type: bool
  • required: False
• input
  • input dataset option
    • see io plugins
  • type: map
  • allowempty: True
  • required: True
  • options(map):
    • type
      • input type
      • type: str
      • required: True
• output
  • output dataset option
    • see io plugins
  • type: map
  • allowempty: True
  • required: True
  • options(map):
    • type
      • input type
      • type: str
      • required: True