Simple Configurations

Below is a list of some simple configurations that can be run with copy/paste config.toml settings in Driverless AI GUI.

Get a quick Final Model: no Genetic Algorithm no Ensembling

These settings can be copy pasted in the Toml editor in the Expert Settings. The experiment preview can be checked to make sure the changes have taken effect. The Toml editor of a completed experiment will also list them at the end of the experiment.

Toml editor

enable_genetic_algorithm = "off"
fixed_ensemble_level = 0

Use Original Features With Genetic Algorithm

This example does no transformations on numeric features and only a single simple encoding on categorical features, i.e. no interactions, target-encoding, dates, text, etc. It only does model selection and tuning via GA.

The examples can be copy pasted in the Toml editor in the Expert Settings. The experiment preview gets modified and can be inspected to confirm the changes have taken effect.

1. The example applies only identity or original transformation on numeric columns and Frequent Transformer on integer and categorical columns, i.e it does not do feature engineering or feature interactions (consider mutation_mode = “full” if set interaction depth >1). It does not drop any features form the original set but you may still not see some features if the modeling algorithm does not use it say for splitting while fitting the model)

   Toml editor

   included_transformers = ["OriginalTransformer","FrequentTransformer"]
   max_feature_interaction_depth = 1
   no_drop_features = true
   

2. This applies identity transformer on numeric features and OneHotEncoding transformation on categorical features.

   Toml editor

   included_transformers = ["OriginalTransformer","OneHotEncodingTransformer"]
   max_feature_interaction_depth = 1
   no_drop_features = true
   

Build models with your choice of algorithm and parameters

These settings can be copy pasted in the Add to config.toml via toml string under the Expert Experiment settings of an experiment. Always check the Driverless preview to make sure the changes have taken effect before launching the experiment. The Scores tab can be used to inspect the built model.

1. This example builds a single GBM model with 2 folds cross validation and user provided parameters with no genetic algorithm.

   Add to config.toml via toml string

   ""  included_models = ['XGBOOSTGBM']\n
       params_xgboost = "{'max_depth': 2, 'max_leaves': 4, 'n_estimators': 50, 'learning_rate': 0.03}"\n
       fixed_num_folds = 2 \n
       feature_brain_level = 0 \n
       enable_genetic_algorithm = "off" \n
   ""
   

2. This example builds a single TensorFlow model on original numeric features with user defined parameters. The model is allowed to predict outside the range of the numeric target column. The model is evaluated with a 4 fold cross validation scheme. Mojo creation, pipeline visualization and genetic algorithm is turned off. Experiment logs can be viewed to verify the parameter used by the TensorFlow model.

   Add to config.toml via toml string

   ""  included_models = ["TensorFlowModel"] \n
       included_transformers = ["OriginalTransformer"] \n
       fixed_ensemble_level = 1 \n
       fixed_num_folds = 4 \n
       params_tensorflow = "{'batch_size': 4096, 'epochs': 100, 'hidden': [1000, 1000]}" \n
       target_transformer = "identity_noclip" \n
       make_mojo_scoring_pipeline = "off" \n
       make_pipeline_visualization = "off" \n
       enable_genetic_algorithm = "off" \n
   ""
   

3. This example builds LightGBM models. During genetic algorithm, it does feature engineering and will do model tuning by toggling other params not set by the user.The Scores tab can be used to inspect the built models.

   Add to config.toml via toml string

   ""  included_models = ["LightGBMModel"] \n
       params_lightgbm = "{ 'n_estimators': 100, 'max_leaves': 64, 'random_state': 1234}" \n
   ""
   

Disable leakage checks, shift detection, mojo creation and ensembling

These settings can be copy pasted in the Toml editor in the Expert Settings. The experiment preview can be checked to make sure the changes have taken effect. The Toml editor of a completed experiment will also list them at the end of the experiment.

Leakages or shift detection can be turned off if user is aware that the data is good or when building a TensorFlow model. These settings can also be applied if user knows the data well and is only interested in the DAI feature engineering on a single model with no mojo creation. Note that num_folds can be set to zero if an external validation set is provided.

Toml editor

check_distribution_shift = "off"
check_leakage = "off"
make_mojo_scoring_pipeline = "off"
fixed_ensemble_level = 0
fixed_num_folds = 0