ClickHouse/docs/zh/guides/apply-catboost-model.md

242 lines
6.6 KiB
Markdown
Raw Normal View History

2020-04-03 13:23:32 +00:00
---
machine_translated: true
machine_translated_rev: 72537a2d527c63c07aa5d2361a8829f3895cf2bd
toc_priority: 41
toc_title: "\u5E94\u7528CatBoost\u6A21\u578B"
2020-04-03 13:23:32 +00:00
---
# 在ClickHouse中应用Catboost模型 {#applying-catboost-model-in-clickhouse}
2020-04-03 13:23:32 +00:00
2020-09-28 08:12:14 +00:00
[CatBoost](https://catboost.ai) 是一个用于机器学习的免费开源梯度提升开发库 [Yandex](https://yandex.com/company/) 。
通过这篇指导您将学会如何将预先从SQL推理出的运行模型作为训练好的模型应用到ClickHouse中去。
2020-04-03 13:23:32 +00:00
在ClickHouse中应用CatBoost模型:
2020-04-03 13:23:32 +00:00
1. [创建表](#create-table).
2. [将数据插入到表中](#insert-data-to-table).
2020-09-07 07:00:47 +00:00
3. [将CatBoost集成到ClickHouse中](#integrate-catboost-into-clickhouse) (可选步骤)。
4. [从SQL运行模型推理](#run-model-inference).
2020-04-03 13:23:32 +00:00
2020-09-28 08:12:14 +00:00
有关训练CatBoost模型的详细信息请参阅 [训练和使用模型](https://catboost.ai/docs/features/training.html#training).
2020-04-03 13:23:32 +00:00
## 先决条件 {#prerequisites}
2020-04-03 13:23:32 +00:00
2020-09-28 08:12:14 +00:00
请先安装好 [Docker](https://docs.docker.com/install/)。
2020-04-03 13:23:32 +00:00
!!! note "注"
[Docker](https://www.docker.com) 是一个软件平台允许您创建容器将CatBoost和ClickHouse安装与系统的其余部分隔离。
2020-04-03 13:23:32 +00:00
在应用CatBoost模型之前:
2020-04-03 13:23:32 +00:00
2020-09-28 08:12:14 +00:00
**1.** 从容器仓库拉取docker映像 (https://hub.docker.com/r/yandex/tutorial-catboost-clickhouse) :
2020-04-03 13:23:32 +00:00
``` bash
$ docker pull yandex/tutorial-catboost-clickhouse
```
此Docker映像包含运行CatBoost和ClickHouse所需的所有内容代码、运行时、库、环境变量和配置文件。
2020-04-03 13:23:32 +00:00
**2.** 确保已成功拉取Docker映像:
2020-04-03 13:23:32 +00:00
``` bash
$ docker image ls
REPOSITORY TAG IMAGE ID CREATED SIZE
yandex/tutorial-catboost-clickhouse latest 622e4d17945b 22 hours ago 1.37GB
```
**3.** 基于此映像启动一个Docker容器:
2020-04-03 13:23:32 +00:00
``` bash
$ docker run -it -p 8888:8888 yandex/tutorial-catboost-clickhouse
```
## 1. 创建表 {#create-table}
2020-04-03 13:23:32 +00:00
为训练样本创建ClickHouse表:
2020-04-03 13:23:32 +00:00
**1.** 在交互模式下启动ClickHouse控制台客户端:
2020-04-03 13:23:32 +00:00
``` bash
$ clickhouse client
```
!!! note "注"
ClickHouse服务器已经在Docker容器内运行。
2020-04-03 13:23:32 +00:00
**2.** 使用以下命令创建表:
2020-04-03 13:23:32 +00:00
``` sql
:) CREATE TABLE amazon_train
(
date Date MATERIALIZED today(),
ACTION UInt8,
RESOURCE UInt32,
MGR_ID UInt32,
ROLE_ROLLUP_1 UInt32,
ROLE_ROLLUP_2 UInt32,
ROLE_DEPTNAME UInt32,
ROLE_TITLE UInt32,
ROLE_FAMILY_DESC UInt32,
ROLE_FAMILY UInt32,
ROLE_CODE UInt32
)
ENGINE = MergeTree ORDER BY date
```
**3.** 从ClickHouse控制台客户端退出:
2020-04-03 13:23:32 +00:00
``` sql
:) exit
```
## 2. 将数据插入到表中 {#insert-data-to-table}
2020-04-03 13:23:32 +00:00
插入数据:
2020-04-03 13:23:32 +00:00
**1.** 运行以下命令:
2020-04-03 13:23:32 +00:00
``` bash
$ clickhouse client --host 127.0.0.1 --query 'INSERT INTO amazon_train FORMAT CSVWithNames' < ~/amazon/train.csv
```
**2.** 在交互模式下启动ClickHouse控制台客户端:
2020-04-03 13:23:32 +00:00
``` bash
$ clickhouse client
```
**3.** 确保数据已上传:
2020-04-03 13:23:32 +00:00
``` sql
:) SELECT count() FROM amazon_train
SELECT count()
FROM amazon_train
+-count()-+
| 65538 |
+-------+
2020-04-03 13:23:32 +00:00
```
2020-09-07 07:00:47 +00:00
## 3. 将CatBoost集成到ClickHouse中 {#integrate-catboost-into-clickhouse}
2020-04-03 13:23:32 +00:00
!!! note "注"
**可选步骤。** Docker映像包含运行CatBoost和ClickHouse所需的所有内容。
2020-04-03 13:23:32 +00:00
2020-09-07 07:00:47 +00:00
CatBoost集成到ClickHouse步骤:
2020-04-03 13:23:32 +00:00
2020-09-28 08:12:14 +00:00
**1.** 构建测试库文件。
2020-04-03 13:23:32 +00:00
2020-09-28 08:12:14 +00:00
测试CatBoost模型的最快方法是编译 `libcatboostmodel.<so|dll|dylib>` 库文件. 有关如何构建库文件的详细信息,请参阅 [CatBoost文件](https://catboost.ai/docs/concepts/c-plus-plus-api_dynamic-c-pluplus-wrapper.html).
2020-04-03 13:23:32 +00:00
2020-09-28 08:12:14 +00:00
**2.** 任意创建一个新目录, 如 `data` 并将创建的库文件放入其中。 Docker映像已经包含了库 `data/libcatboostmodel.so`.
2020-04-03 13:23:32 +00:00
2020-09-28 08:12:14 +00:00
**3.** 任意创建一个新目录来放配置模型, 如 `models`.
2020-04-03 13:23:32 +00:00
2020-09-28 08:12:14 +00:00
**4.** 任意创建一个模型配置文件,如 `models/amazon_model.xml`.
2020-04-03 13:23:32 +00:00
**5.** 描述模型配置:
2020-04-03 13:23:32 +00:00
``` xml
<models>
<model>
<!-- Model type. Now catboost only. -->
<type>catboost</type>
<!-- Model name. -->
<name>amazon</name>
<!-- Path to trained model. -->
<path>/home/catboost/tutorial/catboost_model.bin</path>
<!-- Update interval. -->
<lifetime>0</lifetime>
</model>
</models>
```
2020-09-28 08:12:14 +00:00
**6.** 将CatBoost库文件的路径和模型配置添加到ClickHouse配置:
2020-04-03 13:23:32 +00:00
``` xml
<!-- File etc/clickhouse-server/config.d/models_config.xml. -->
<catboost_dynamic_library_path>/home/catboost/data/libcatboostmodel.so</catboost_dynamic_library_path>
<models_config>/home/catboost/models/*_model.xml</models_config>
```
2020-09-28 08:12:14 +00:00
## 4. 运行从SQL推理的模型 {#run-model-inference}
2020-04-03 13:23:32 +00:00
2020-09-28 08:12:14 +00:00
测试模型是否正常运行ClickHouse客户端 `$ clickhouse client`.
2020-04-03 13:23:32 +00:00
2020-09-28 08:12:14 +00:00
让我们确保模型能正常工作:
2020-04-03 13:23:32 +00:00
``` sql
:) SELECT
modelEvaluate('amazon',
RESOURCE,
MGR_ID,
ROLE_ROLLUP_1,
ROLE_ROLLUP_2,
ROLE_DEPTNAME,
ROLE_TITLE,
ROLE_FAMILY_DESC,
ROLE_FAMILY,
ROLE_CODE) > 0 AS prediction,
ACTION AS target
FROM amazon_train
LIMIT 10
2020-09-28 08:12:14 +00:00
```
2020-04-03 13:23:32 +00:00
!!! note "注"
2020-09-28 08:12:14 +00:00
函数 [modelEvaluate](../sql-reference/functions/other-functions.md#function-modelevaluate) 返回带有多类模型的每类原始预测的元组。
2020-04-03 13:23:32 +00:00
让我们预测一下:
2020-04-03 13:23:32 +00:00
``` sql
:) SELECT
modelEvaluate('amazon',
RESOURCE,
MGR_ID,
ROLE_ROLLUP_1,
ROLE_ROLLUP_2,
ROLE_DEPTNAME,
ROLE_TITLE,
ROLE_FAMILY_DESC,
ROLE_FAMILY,
ROLE_CODE) AS prediction,
1. / (1 + exp(-prediction)) AS probability,
ACTION AS target
FROM amazon_train
LIMIT 10
```
!!! note "注"
2020-09-28 08:12:14 +00:00
查看函数说明 [exp()](../sql-reference/functions/math-functions.md) 。
2020-04-03 13:23:32 +00:00
让我们计算样本的LogLoss:
2020-04-03 13:23:32 +00:00
``` sql
:) SELECT -avg(tg * log(prob) + (1 - tg) * log(1 - prob)) AS logloss
FROM
(
SELECT
modelEvaluate('amazon',
RESOURCE,
MGR_ID,
ROLE_ROLLUP_1,
ROLE_ROLLUP_2,
ROLE_DEPTNAME,
ROLE_TITLE,
ROLE_FAMILY_DESC,
ROLE_FAMILY,
ROLE_CODE) AS prediction,
1. / (1. + exp(-prediction)) AS prob,
ACTION AS tg
FROM amazon_train
)
```
!!! note "注"
2020-09-28 08:12:14 +00:00
查看函数说明 [avg()](../sql-reference/aggregate-functions/reference.md#agg_function-avg) 和 [log()](../sql-reference/functions/math-functions.md) 。
2020-04-03 13:23:32 +00:00
[原始文章](https://clickhouse.tech/docs/en/guides/apply_catboost_model/) <!--hide-->