ClickHouse/docs/en/sql-reference/functions/machine-learning-functions.md
2021-07-29 18:20:55 +03:00

97 lines
3.3 KiB
Markdown

---
toc_priority: 64
toc_title: Machine Learning
---
# Machine Learning Functions {#machine-learning-functions}
## evalMLMethod {#machine_learning_methods-evalmlmethod}
Prediction using fitted regression models uses `evalMLMethod` function. See link in `linearRegression`.
## stochasticLinearRegression {#stochastic-linear-regression}
The [stochasticLinearRegression](../../sql-reference/aggregate-functions/reference/stochasticlinearregression.md#agg_functions-stochasticlinearregression) aggregate function implements stochastic gradient descent method using linear model and MSE loss function. Uses `evalMLMethod` to predict on new data.
## stochasticLogisticRegression {#stochastic-logistic-regression}
The [stochasticLogisticRegression](../../sql-reference/aggregate-functions/reference/stochasticlogisticregression.md#agg_functions-stochasticlogisticregression) aggregate function implements stochastic gradient descent method for binary classification problem. Uses `evalMLMethod` to predict on new data.
## bayesAB {#bayesab}
Compares test groups (variants) and calculates for each group the probability to be the best one. The first group is used as a control group.
**Syntax**
``` sql
bayesAB(distribution_name, higher_is_better, variant_names, x, y)
```
**Arguments**
- `distribution_name` — Name of the probability distribution. [String](../../sql-reference/data-types/string.md). Possible values:
- `beta` for [Beta distribution](https://en.wikipedia.org/wiki/Beta_distribution)
- `gamma` for [Gamma distribution](https://en.wikipedia.org/wiki/Gamma_distribution)
- `higher_is_better` — Boolean flag. [Boolean](../../sql-reference/data-types/boolean.md). Possible values:
- `0` — lower values are considered to be better than higher
- `1` — higher values are considered to be better than lower
- `variant_names` — Variant names. [Array](../../sql-reference/data-types/array.md)([String](../../sql-reference/data-types/string.md)).
- `x` — Numbers of tests for the corresponding variants. [Array](../../sql-reference/data-types/array.md)([Float64](../../sql-reference/data-types/float.md)).
- `y` — Numbers of successful tests for the corresponding variants. [Array](../../sql-reference/data-types/array.md)([Float64](../../sql-reference/data-types/float.md)).
!!! note "Note"
All three arrays must have the same size. All `x` and `y` values must be non-negative constant numbers. `y` cannot be larger than `x`.
**Returned values**
For each variant the function calculates:
- `beats_control` — long-term probability to out-perform the first (control) variant
- `to_be_best` — long-term probability to out-perform all other variants
Type: JSON.
**Example**
Query:
``` sql
SELECT bayesAB('beta', 1, ['Control', 'A', 'B'], [3000., 3000., 3000.], [100., 90., 110.]) FORMAT PrettySpace;
```
Result:
``` text
{
"data":[
{
"variant_name":"Control",
"x":3000,
"y":100,
"beats_control":0,
"to_be_best":0.22619
},
{
"variant_name":"A",
"x":3000,
"y":90,
"beats_control":0.23469,
"to_be_best":0.04671
},
{
"variant_name":"B",
"x":3000,
"y":110,
"beats_control":0.7580899999999999,
"to_be_best":0.7271
}
]
}
```