Skip to content

Latest commit

 

History

History
60 lines (46 loc) · 1.53 KB

README.md

File metadata and controls

60 lines (46 loc) · 1.53 KB
Raw

FAVEL ML Service

Configuration

MLApproches tag in the configuration file favel.conf has three variable method, parameters, and normalizer.

[MLApproches]
method = GradientBoostingClassifier
parameters = default
normalizer = default

method

The machine learning (ML) algorithm has to be specified to the method. The algorithm name must be the sklearn module.

method = GradientBoostingClassifier

parameters

default is used to run an experiment on the default ML algorithm parameters.

parameters = default

To set specified parameters of ML algorithm. The parameters should be passed in JSON format.

parameters = {'n_estimators': 100, 'learning_rate': 1.0, 'max_depth': 1, 'random_state': 0}

For optimization of ML Algorithm parameters. Parameters name with range is specified.

parameters = [
            {'name':'n_estimators',  'range':(1, 3)},
            {'name':'learning_rate', 'range':(1.0,2.0)},
            {'name': 'warm_start', 'range':[True, False]}
            ]

normalizer

To normalize data used for ML. Any one of these normalizers should be specified in the config file.

normalizer = default
normalizer = Normalizer
normalizer = MinMaxScaler
normalizer = StandardScaler
normalizer = MaxAbsScaler
normalizer = RobustScaler

Outputs

  • Classification Report.xlsx shows the performance of the ML model for each class.
  • classifier.pkl trained ml model dump in pickle format.
  • predicate_le.pkl label encoding for predicates dump in pickle format.