机器学习_kedro框架使用简单示意

简介及安装包

kedro用来构建可复用，易维护，模块化的机器学习代码。相比于Notebook的超级灵活性,便于探索数据和算法， Kedro 定位于解决版本控制，可复用性，文档，单元测试，部署等工程方面的问题。

python -m pip install --upgrade pip
pip install kedro
pip install kedro-viz
# check 
kedro info

一、创建项目

项目创建

kedro new
# 并输入
iris_demo

生成项目的目录如下

(base) [~iris_demo]$ tree
.
├── conf
│   ├── base
│   │   ├── catalog.yml
│   │   ├── logging.yml
│   │   └── parameters.yml
│   ├── local
│   │   └── credentials.yml
│   └── README.md
├── data
│   ├── 01_raw
│   ├── 02_intermediate
│   ├── 03_primary
│   ├── 04_feature
│   ├── 05_model_input
│   ├── 06_models
│   ├── 07_model_output
│   └── 08_reporting
├── docs
│   └── source
│       ├── conf.py
│       └── index.rst
├── logs
├── notebooks
├── pyproject.toml
├── README.md
├── setup.cfg
└── src
├── iris_demo
│   ├── init.py
│   ├── main.py
│   ├── pipeline_registry.py
│   ├── pipelines
│   │   └── init.py
│   └── settings.py
├── requirements.txt
├── setup.py
└── tests
├── init.py
├── pipelines
│   └── init.py
└── test_run.py

• 目录作用说明

iris_demo        # Parent directory of the template
├── conf            # Project configuration files
├── data            # 项目本地文件 (not committed to version control)
├── docs            # 项目文档  Project documentation
├── logs            # 项目日志  Project output logs (not committed to version control)
├── notebooks       # 项目的一些notebooks Project related Jupyter notebooks (can be used for experimental code before moving the code to src)
├── README.md       # Project README
├── setup.cfg       # Configuration options for `pytest` when doing `kedro test` and for the `isort` utility when doing `kedro lint`
└── src             # Project source code

2.1 data 完善

直接将sklearn中的iris数据导出

cd data/05_model_input
# 进入python
python

生成 iris.csv 文件

from sklearn.datasets import load_iris
import pandas as pd

iris = load_iris()
df = pd.DataFrame(iris.data, columns=[i[:-5].replace(' ', '_') for i in iris.feature_names])
df['target'] = iris.target
df.to_csv('iris.csv', index=False)

2.2 src 完善

cd src/iris_demo/pipelines
mkdir training_pipline
cd training_pipline
touch train_func_node.py
touch train_pip.py

• train_func_node.py 将训练脚本写成function 然后用于后续封装

# train_func_node.py 
# python3
# Create date: 2022-09-05
# Author: Scc_hy
# Func: 模型训练
# ===============================================================================
import logging
import pandas as pd
from sklearn.linear_model import LogisticRegression
from sklearn.metrics import f1_score
from sklearn.model_selection import train_test_split


def split_data(data, parameters):
    X = data[parameters["features"]]
    y = data[parameters["target"]]
    X_train, X_test, y_train, y_test = train_test_split(
        X, y, test_size=parameters["test_size"], random_state=parameters["random_state"]
    )
    return X_train, X_test, y_train, y_test


def train_model(X_train, y_train):
    regressor = LogisticRegression()
    regressor.fit(X_train, y_train.values.ravel())
    return regressor


def evaluate_model(estimator, X_test, y_test):
    y_pred = estimator.predict(X_test)
    score = f1_score(y_test.values.ravel(), y_pred.ravel(), average='macro')
    logger = logging.getLogger(__name__)
    logger.info(f"[ valid ] f1-score {score:.3f}")

• train_pip.py 将train_func_node.py 中的方法封装成node, 然后构建一个pipeline
  • 如果输入不是上面的node的输出，则需要在conf/base/catalog.yml 或者 conf/base/parameters.yml 中定义的
    • parameters.yml 中 params:model_options
    • conf/base/catalog.yml 中 iris_data
  • 对于模型以及一些中间数据，我们可以在 conf/base/catalog.yml 中设置添加

# python3
# Create date: 2022-09-05
# Author: Scc_hy
# Func: 模型训练 pipeline
# ===============================================================================

from kedro.pipeline import pipeline, node
from .train_func_node import evaluate_model, split_data, train_model


def create_pipeline(**kwargs):
    return pipeline(
        [
            node(
                func=split_data,
                inputs=["irir_data", "params:model_options"],
                outputs=["X_train", "X_test", "y_train", "y_test"],
                name="split_data_node",
            ),
            node(
                func=train_model,
                inputs=["X_train", "y_train"],
                outputs="logistic_model_v1",
                name="train_model_node",
            ),
            node(
                func=evaluate_model,
                inputs=["logistic_model_v1", "X_test", "y_test"],
                outputs=None,
                name="evaluate_model_node",
            ),
        ]
    )

完善 iris-demo\src\iris_demo\pipeline_registry.py

"""Project pipelines."""
from typing import Dict
from iris_demo.pipelines.training_pipline.train_pip import create_pipeline
from kedro.pipeline import Pipeline, pipeline


def register_pipelines() -> Dict[str, Pipeline]:
    """Register the project's pipeline.

    Returns:
        A mapping from a pipeline name to a ``Pipeline`` object.

    """
    data_science_pipeline = create_pipeline()
    return {
        "__default__": data_science_pipeline,
        "ds": data_science_pipeline,
    }

2.3 conf 完善

文件作用说明

# conf/base
catalog.yml # with the file paths and load/save configuration required for different datasets
logging.yml # Uses Python’s default logging library to set up logging
parameters.yml # Allows you to define parameters for machine learning experiments e.g. train / test split and number of iterations


# settings that should not be shared
# the contents of conf/local/ is ignored by git
# conf/local 
credentials.yml

完善 catalog.yml

在 conf/base/catalog.yml 中定义所有的数据

irir_data:
  type: pandas.CSVDataSet
  filepath: data/05_model_input/iris.csv

logistic_model_v1:
  type: pickle.PickleDataSet
  filepath: data/06_models/logistic_model_v1.pickle
  versioned: true

完善 parameters.yml

model_options:
  test_size: 0.2
  random_state: 66
  features:
    - sepal_length
    - sepal_width
    - petal_length
    - petal_width
  target: 
    - target

2.4 运行项目

kedro run

二、创建训练视图

kedro viz --port 2022

参考

https://kedro.readthedocs.io/en/stable/get_started/example_project.html#run-the-example-project