---
title: Machine Learning Functions reference
meta:
    description: Functions for machine learning.
headingMaxLevels: 2
---

# Machine Learning functions

The following functions are used for machine learning tasks, primarily for training and applying linear and logistic regression models.

## evalMLMethod

Applies a previously trained machine learning model to new input features to generate a prediction. This function is used to make inferences after a model has been trained using an aggregate function like `stochasticLinearRegressionState` or `stochasticLogisticRegressionState`.

### Syntax

```sql
evalMLMethod(model_state, feature1, feature2, ...)
```

### Arguments

- `model_state`: `AggregateFunctionState`. The internal state of a trained machine learning model, typically obtained from an aggregate function.
- `feature1`, `feature2`, ...: `Numeric`. One or more input features for which to make a prediction.

### Returns

The predicted numerical value from the model. `Float64`.

## stochasticLinearRegression

An aggregate function that trains a linear regression model using the stochastic gradient descent method. It aims to minimize the mean squared error (MSE) loss function. The result of this aggregate function is a model state that can then be used with `evalMLMethod` for predictions.

### Syntax

```sql
stochasticLinearRegression(learning_rate, L1, L2, no_bias, target, feature1, feature2, ...)
```

### Arguments

- `learning_rate`: `Float64`. The step size at each iteration while moving toward a minimum of the loss function.
- `L1`: `Float64`. The L1 regularization parameter.
- `L2`: `Float64`. The L2 regularization parameter.
- `no_bias`: `UInt8`. A flag (0 or 1) to indicate whether to exclude the bias term from the model.
- `target`: `Numeric`. The dependent variable (label) for training.
- `feature1`, `feature2`, ...: `Numeric`. One or more independent variables (features) for training.

### Returns

An internal state representing the trained linear regression model. `AggregateFunctionState`.

## stochasticLogisticRegression

An aggregate function that trains a logistic regression model using the stochastic gradient descent method, primarily for binary classification problems. The output is a model state that can be used with `evalMLMethod` to predict class probabilities.

### Syntax

```sql
stochasticLogisticRegression(learning_rate, L1, L2, no_bias, target, feature1, feature2, ...)
```

### Arguments

- `learning_rate`: `Float64`. The step size at each iteration while moving toward a minimum of the loss function.
- `L1`: `Float64`. The L1 regularization parameter.
- `L2`: `Float64`. The L2 regularization parameter.
- `no_bias`: `UInt8`. A flag (0 or 1) to indicate whether to exclude the bias term from the model.
- `target`: `Numeric`. The dependent variable (label) for training, typically 0 or 1 for binary classification.
- `feature1`, `feature2`, ...: `Numeric`. One or more independent variables (features) for training.

### Returns

An internal state representing the trained logistic regression model. `AggregateFunctionState`.