Skip to main content

Documentation Index

Fetch the complete documentation index at: https://private-7c7dfe99-mintlify-replace-private-preview-badge-man.mintlify.app/llms.txt

Use this file to discover all available pages before exploring further.

​
stochasticLogisticRegression

Introduced in: v20.1.0 This function implements stochastic logistic regression. It can be used for binary classification problem, supports the same custom parameters as stochasticLinearRegression and works the same way. Usage The function is used in two steps:
  1. Fitting
For fitting a query like this can be used: 
CREATE TABLE IF NOT EXISTS train_data
(
    param1 Float64,
    param2 Float64,
    target Float64
) ENGINE = Memory;

CREATE TABLE your_model ENGINE = Memory AS SELECT
stochasticLogisticRegression(0.1, 0.0, 5, 'SGD')(target, x1, x2)
AS state FROM train_data;
Here, we also need to insert data into the train_data table. The number of parameters is not fixed, it depends only on the number of arguments passed into logisticRegressionState. They all must be numeric values. Note that the column with target value (which we would like to learn to predict) is inserted as the first argument. Predicted labels have to be in [-1, 1].
  1. Predicting
Using saved state we can predict the probability of an object having label 1. 
WITH (SELECT state FROM your_model) AS model SELECT
evalMLMethod(model, param1, param2) FROM test_data
The query will return a column of probabilities. Note that first argument of evalMLMethod is an AggregateFunctionState object, next are columns of features. We can also set a bound of probability, which assigns elements to different labels. 
SELECT result < 1.1 AND result > 0.5 FROM
(WITH (SELECT state FROM your_model) AS model SELECT
evalMLMethod(model, param1, param2) AS result FROM test_data)
Then the result will be labels. test_data is a table like train_data but may not contain target value. Syntax 
stochasticLogisticRegression([learning_rate, l2_regularization_coef, mini_batch_size, method])(target, x1, x2, ...)
Arguments
  • learning_rate — The coefficient on step length when gradient descent step is performed. A learning rate that is too big may cause infinite weights of the model. Default is 0.00001. Float64
  • l2_regularization_coef — L2 regularization coefficient which may help to prevent overfitting. Default is 0.1. Float64
  • mini_batch_size — Sets the number of elements which gradients will be computed and summed to perform one step of gradient descent. Pure stochastic descent uses one element, however having small batches (about 10 elements) makes gradient steps more stable. Default is 15. UInt64
  • method — Method for updating weights: Adam (by default), SGD, Momentum, Nesterov. Momentum and Nesterov require a little bit more computations and memory, however they happen to be useful in terms of speed of convergence and stability of stochastic gradient methods. String
  • target — Target binary classification labels. Must be in range [-1, 1]. Float
  • x1, x2, ... — Feature values (independent variables). All must be numeric. Float
Returned value Returns the trained logistic regression model weights. Use evalMLMethod for predictions which returns probabilities of object having label 1. Array(Float64) Examples Training a model
Query
CREATE TABLE your_model
ENGINE = MergeTree
ORDER BY tuple()
AS SELECT
stochasticLogisticRegressionState(1.0, 1.0, 10, 'SGD')(target, x1, x2)
AS state FROM train_data
Response
Saves trained model state to table
Making predictions
Query
WITH (SELECT state FROM your_model) AS model
SELECT
evalMLMethod(model, x1, x2)
FROM test_data
Response
Returns probability values for test data
Classification with threshold
Query
SELECT result < 1.1 AND result > 0.5
FROM (
WITH (SELECT state FROM your_model) AS model SELECT
evalMLMethod(model, x1, x2) AS result FROM test_data)
Response
Returns binary classification labels using probability threshold
See Also