Flexible generation

Generators in MOSTLY AI have flexible generation enabled by default. You can enable or disable flexible generation for each table in a generator. You can do so before you train a generator. Once a generator is trained, you cannot change its flexible generation configuration.

Even if enabled, the use of rebalancing, generation temperature options (Temperature and Top P), and fairness is optional.

If you want to restrict your models and disable the options to rebalance columns, set the generation temperature, or use fairness, you can disable flexible generation before starting training.

Why flexible generation?

Flexible generation enables a generator to rebalance data, change the generation temperature, and use fairness.

• With rebalancing, you can rebalance your datasets in a highly correlated manner. For a simplified example, if you decide to increase the percentage of women in your dataset, correlated features such as female first names, respective average income, respective job titles, will also be impacted.
• With Temperature and Top P, you can control how creatively or conservatively your AI models generate synthetic data.
• With fairness, you can generate statistical parity synthetic data and define which categorical columns (for example, income) should be generated fairly based on which other sensitive columns (for example, sex, race, age, and so on).

You can use flexible generation for scenarios such as fraud detection, statistical parity synthetic data, what-if scenarios, testing of an increase in less likely outcomes or decrease in rather likely outcomes.

Enable flexible generation

Flexible generation is on by default for each table in a generator.

config = {
    "name": "US census",
    "tables": [
        {
            "data": "https://mostly.ai/docs/datasets/us-census-income.csv",
            "name": "census",
            "modelConfiguration": {
                "enableFlexibleGeneration": True,
            }
        }
    ]
}