Auto ML

The Auto ML feature in AryaXAI enables users to generate their own models if they do not already possess one. In this section, you can access all the models that we've uploaded or create a new model. You can easily  activate or deactivate any model in this section.

Upon creating a new project, AryaXAI automatically trains a default AutoML model, 'XGBoost_default', for default prediction and explainability.

Users can easily train a model using AryaXAI's built-in modelling techniques, which include:

• XGBoost
• LGBoost
• CatBoost
• RandomForest
• SGD (Stochastic Gradient Descent)
• Logistic Regression
• Linear Regression
• GaussianNaiveBayes

Easily fine-tune these models and adjust the hyperparameters according to your requirements.

To train a model and retrieve a list of all trained models within the project:

# train model
project.train_model(model_type='RandomForest')

# Train model. This Trains Another version Model with Current config (If config not passed)
project.train_model()

Help function to train a model:

help(project.train_model)

View available models to train

project.available_models()

See uploaded model info

project.models()

#Instance_type is used to specify what kind of serverless instance type you want this model training function to run. The options are 'small', and 'large'.

#feature_encodings : you can change the encoding type for each feature. Available options are "labelencode , countencode"

feature_exclude: You can pass new exclusions from SDK.

NOTE: Ensure your new exclusion list is updated. In the SDK mode, we use your current feature exclusion list as the revised one and overwrite the previous list.

You can modify model hyperparameters as well.

project.model_parameters()

Good to know:

• What if I want to train the model only on a sample of my training data?

You can define 'sample_percentage' in data_config. This uses samples of the training for model training.

• How can I customize the base sample for the SHAP?

You can define 'explainability_sample_percentage' in data_config. This uses a sample of the training for running shapely values. Use around 10K or 10% of training data as:

data_config= {
    "tags": ['Training'],
    "drop_duplicate_uid": True,
    "feature_exclude": feature_exclude,
    'sample_percentage': 0.1,
    'explainability_sample_percentage': 0.1
}

Delete a trained model:

project.remove_model()