MetaGPT

中文

English

中文

English

Appearance

使用工具进行机器学习建模 ​

工具批量使用 ​

任务 ​

使用 DataInterpreter 对 titanic 数据集进行建模预测，并在数据预处理和特征工程阶段批量调用工具（已经预先创建了多个 data_preprocess 和 feature_engineering 类型的工具，位于目录metagpt/tools/libs下）。

代码 ​

bash
python examples/di/machine_learning_with_tools.py
python examples/di/machine_learning_with_tools.py

运行结果 ​

数据预处理代码（其中包含 3 个工具的使用：FillMissingValue、StandardScale 和 OneHotEncode）如下：

python
from metagpt.tools.libs.data_preprocess import FillMissingValue, StandardScale, OneHotEncode

fill_age = FillMissingValue(features=['Age'], strategy='median')
fill_embarked = FillMissingValue(features=['Embarked'], strategy='most_frequent')

train_data = fill_age.fit_transform(train_data)
train_data = fill_embarked.fit_transform(train_data)

if 'Cabin' in train_data.columns:
    train_data.drop('Cabin', axis=1, inplace=True)

numerical_features = ['Age', 'SibSp', 'Parch', 'Fare']

scale = StandardScale(features=numerical_features)
scaled_features = scale.fit_transform(train_data)

for feature in numerical_features:
    train_data[feature] = scaled_features[feature]

categorical_features = ['Sex', 'Embarked']
one_hot = OneHotEncode(features=categorical_features)
train_data = one_hot.fit_transform(train_data)

non_informative_columns = ['PassengerId', 'Name', 'Ticket']
train_data.drop(columns=non_informative_columns, axis=1, inplace=True, errors='ignore')
train_data.head()
from metagpt.tools.libs.data_preprocess import FillMissingValue, StandardScale, OneHotEncode

fill_age = FillMissingValue(features=['Age'], strategy='median')
fill_embarked = FillMissingValue(features=['Embarked'], strategy='most_frequent')

train_data = fill_age.fit_transform(train_data)
train_data = fill_embarked.fit_transform(train_data)

if 'Cabin' in train_data.columns:
    train_data.drop('Cabin', axis=1, inplace=True)

numerical_features = ['Age', 'SibSp', 'Parch', 'Fare']

scale = StandardScale(features=numerical_features)
scaled_features = scale.fit_transform(train_data)

for feature in numerical_features:
    train_data[feature] = scaled_features[feature]

categorical_features = ['Sex', 'Embarked']
one_hot = OneHotEncode(features=categorical_features)
train_data = one_hot.fit_transform(train_data)

non_informative_columns = ['PassengerId', 'Name', 'Ticket']
train_data.drop(columns=non_informative_columns, axis=1, inplace=True, errors='ignore')
train_data.head()

特征工程代码（其中包含 1 个工具的使用：PolynomialExpansion）如下：

python
from metagpt.tools.libs.feature_engineering import PolynomialExpansion
numeric_features = ['Pclass', 'Age', 'SibSp', 'Parch', 'Fare']
poly_expansion = PolynomialExpansion(cols=numeric_features, label_col='Survived')
train_data = poly_expansion.fit_transform(train_data)
train_data.head()
from metagpt.tools.libs.feature_engineering import PolynomialExpansion
numeric_features = ['Pclass', 'Age', 'SibSp', 'Parch', 'Fare']
poly_expansion = PolynomialExpansion(cols=numeric_features, label_col='Survived')
train_data = poly_expansion.fit_transform(train_data)
train_data.head()

机制解释 ​

工具创建 ​

在 metagpt.tools.libs 目录下，我们预先定义了两类工具：数据预处理（data_preprocess）和特征工程（feature_engineering）。下面以数据预处理任务为例，给出一个简单工具的完整创建指引，包括工具的定义和注册方法。

  • 首先，在 metagpt.tools.libs 目录下创建文件 data_preprocess.py。
  • 在该文件中，您将定义您的数据预处理工具，并利用 @register_tool 装饰器一键完成工具的注册。此外，你也可以通过设置 tags 参数，指定工具的类别。

代码示例：

python
from typing import Literal

import pandas as pd
from sklearn.impute import SimpleImputer

from metagpt.tools.tool_registry import register_tool


TAGS = ["data preprocessing", "machine learning"]

@register_tool(tags=TAGS)
class FillMissingValue:
    """
    Completing missing values with simple strategies.
    """

    def __init__(
        self, features: list, strategy: Literal["mean", "median", "most_frequent", "constant"] = "mean", fill_value=None
    ):
        """
        Initialize self.

        Args:
            features (list): Columns to be processed.
            strategy (Literal["mean", "median", "most_frequent", "constant"], optional): The imputation strategy, notice 'mean' and 'median' can only
                                      be used for numeric features. Defaults to 'mean'.
            fill_value (int, optional): Fill_value is used to replace all occurrences of missing_values.
                                        Defaults to None.
        """
        self.features = features
        self.strategy = strategy
        self.fill_value = fill_value
        self.si = None

    def fit(self, df: pd.DataFrame):
        """
        Fit the FillMissingValue model.

        Args:
            df (pd.DataFrame): The input DataFrame.
        """
        if len(self.features) == 0:
            return
        self.si = SimpleImputer(strategy=self.strategy, fill_value=self.fill_value)
        self.si.fit(df[self.features])

    def transform(self, df: pd.DataFrame) -> pd.DataFrame:
        """
        Transform the input DataFrame with the fitted model.

        Args:
            df (pd.DataFrame): The input DataFrame.

        Returns:
            pd.DataFrame: The transformed DataFrame.
        """
        if len(self.features) == 0:
            return df
        new_df = df.copy()
        new_df[self.features] = self.si.transform(new_df[self.features])
        return new_df

    def fit_transform(self, df: pd.DataFrame) -> pd.DataFrame:
        """
        Fit and transform the input DataFrame.

        Args:
            df (pd.DataFrame): The input DataFrame.

        Returns:
            pd.DataFrame: The transformed DataFrame.
        """
        self.fit(df)
        return self.transform(df)
from typing import Literal

import pandas as pd
from sklearn.impute import SimpleImputer

from metagpt.tools.tool_registry import register_tool


TAGS = ["data preprocessing", "machine learning"]

@register_tool(tags=TAGS)
class FillMissingValue:
    """
    Completing missing values with simple strategies.
    """

    def __init__(
        self, features: list, strategy: Literal["mean", "median", "most_frequent", "constant"] = "mean", fill_value=None
    ):
        """
        Initialize self.

        Args:
            features (list): Columns to be processed.
            strategy (Literal["mean", "median", "most_frequent", "constant"], optional): The imputation strategy, notice 'mean' and 'median' can only
                                      be used for numeric features. Defaults to 'mean'.
            fill_value (int, optional): Fill_value is used to replace all occurrences of missing_values.
                                        Defaults to None.
        """
        self.features = features
        self.strategy = strategy
        self.fill_value = fill_value
        self.si = None

    def fit(self, df: pd.DataFrame):
        """
        Fit the FillMissingValue model.

        Args:
            df (pd.DataFrame): The input DataFrame.
        """
        if len(self.features) == 0:
            return
        self.si = SimpleImputer(strategy=self.strategy, fill_value=self.fill_value)
        self.si.fit(df[self.features])

    def transform(self, df: pd.DataFrame) -> pd.DataFrame:
        """
        Transform the input DataFrame with the fitted model.

        Args:
            df (pd.DataFrame): The input DataFrame.

        Returns:
            pd.DataFrame: The transformed DataFrame.
        """
        if len(self.features) == 0:
            return df
        new_df = df.copy()
        new_df[self.features] = self.si.transform(new_df[self.features])
        return new_df

    def fit_transform(self, df: pd.DataFrame) -> pd.DataFrame:
        """
        Fit and transform the input DataFrame.

        Args:
            df (pd.DataFrame): The input DataFrame.

        Returns:
            pd.DataFrame: The transformed DataFrame.
        """
        self.fit(df)
        return self.transform(df)

在上述示例中，FillMissingValue 类具体实现了缺失值的填充功能，并通过 @register_tool 装饰器注册为数据预处理工具。

注册工具后，系统将自动根据工具函数的注释在 metagpt.tools.schemas 目录下为您生成每个工具生成相应的 schema 文件（如 FillMissingValue.yml），详细描述工具类的结构、方法以及参数类型，确保LLM能够准确理解并使用该工具。

Schema 文件示例（FillMissingValue.yml）：

yaml
type: class
description: Completing missing values with simple strategies.
methods:
  __init__:
    type: function
    description: 'Initialize self. '
    signature:
      '(self, features: ''list'', strategy: "Literal[''mean'', ''median'',
      ''most_frequent'', ''constant'']" = ''mean'', fill_value=None)'
    parameters:
      'Args: features (list): Columns to be processed. strategy (Literal["mean",
      "median", "most_frequent", "constant"], optional): The imputation strategy,
      notice ''mean'' and ''median'' can only be used for numeric features. Defaults
      to ''mean''. fill_value (int, optional): Fill_value is used to replace all occurrences
      of missing_values. Defaults to None.'
  fit:
    type: function
    description: 'Fit a model to be used in subsequent transform. '
    signature: "(self, df: 'pd.DataFrame')"
    parameters: 'Args: df (pd.DataFrame): The input DataFrame.'
  fit_transform:
    type: function
    description: 'Fit and transform the input DataFrame. '
    signature: "(self, df: 'pd.DataFrame') -> 'pd.DataFrame'"
    parameters:
      'Args: df (pd.DataFrame): The input DataFrame. Returns: pd.DataFrame:
      The transformed DataFrame.'
  transform:
    type: function
    description: 'Transform the input DataFrame with the fitted model. '
    signature: "(self, df: 'pd.DataFrame') -> 'pd.DataFrame'"
    parameters:
      'Args: df (pd.DataFrame): The input DataFrame. Returns: pd.DataFrame:
      The transformed DataFrame.'
type: class
description: Completing missing values with simple strategies.
methods:
  __init__:
    type: function
    description: 'Initialize self. '
    signature:
      '(self, features: ''list'', strategy: "Literal[''mean'', ''median'',
      ''most_frequent'', ''constant'']" = ''mean'', fill_value=None)'
    parameters:
      'Args: features (list): Columns to be processed. strategy (Literal["mean",
      "median", "most_frequent", "constant"], optional): The imputation strategy,
      notice ''mean'' and ''median'' can only be used for numeric features. Defaults
      to ''mean''. fill_value (int, optional): Fill_value is used to replace all occurrences
      of missing_values. Defaults to None.'
  fit:
    type: function
    description: 'Fit a model to be used in subsequent transform. '
    signature: "(self, df: 'pd.DataFrame')"
    parameters: 'Args: df (pd.DataFrame): The input DataFrame.'
  fit_transform:
    type: function
    description: 'Fit and transform the input DataFrame. '
    signature: "(self, df: 'pd.DataFrame') -> 'pd.DataFrame'"
    parameters:
      'Args: df (pd.DataFrame): The input DataFrame. Returns: pd.DataFrame:
      The transformed DataFrame.'
  transform:
    type: function
    description: 'Transform the input DataFrame with the fitted model. '
    signature: "(self, df: 'pd.DataFrame') -> 'pd.DataFrame'"
    parameters:
      'Args: df (pd.DataFrame): The input DataFrame. Returns: pd.DataFrame:
      The transformed DataFrame.'

自动工具注册与选择 ​

通过设置DataInterpreter的参数tools，可以人为指定使用哪些工具。如果希望使用所有工具，可以设定tools=["<all>"]。

当 DataInterpreter 启动时，它会自动注册 tools 列表中的所有工具。同时，在任务规划阶段，DataInterpreter 会根据需求为每个任务推荐可用的工具，并挑选出 TOP-k 供 LLM 使用。

工具组合与调用 ​

在执行任务时，一旦可用的 TOP 工具被识别，DataInterpreter 会根据任务需求自动组合并调用其中的有用工具，以形成针对特定任务的代码解决方案。

Released under the MIT License.

Copyright © 2023-present alexanderwu