tsfresh 是一个自动化提取时序特征的库

tsfresh 安装

$ pip install tsfresh

tsfresh 使用步骤

使用tsfresh的使用步骤如下：

前期训练阶段：

1. 数据准备：准备符合 tsfresh 输入格式的数据集
2. 样本抽样：以步长 s 为间隔滑窗抽样
3. 特征生成：对采样样本生成特征，并收集它们
4. 特征选择：收集多个特征下的衍生特征，进行特征选择

后期部署阶段：

1. 数据准备：准备符合 tsfresh 输入格式的数据集
2. 特征选择：对滑窗样本生成特征，并收集它们

tsfresh 数据格式

输入数据格式

• Flat DataFrame
• Stacked DataFrame
• dictionary of flat DataFrame

适合的 API:

• tsfresh.extract_features()
• tsfresh.

Flat DataFrame

Stacked DataFrame

Dictionary of flat DataFrame

{ 
    "x”:
        | id | time | value    |
        |----|------|----------|
        | A  | t1   | x(A, t1) |
        | A  | t2   | x(A, t2) |
        | A  | t3   | x(A, t3) |
        | B  | t1   | x(B, t1) |
        | B  | t2   | x(B, t2) |
        | B  | t3   | x(B, t3) |
  , "y”:
        | id | time | value    |
        |----|------|----------|
        | A  | t1   | y(A, t1) |
        | A  | t2   | y(A, t2) |
        | A  | t3   | y(A, t3) |
        | B  | t1   | y(B, t1) |
        | B  | t2   | y(B, t2) |
        | B  | t3   | y(B, t3) |
}

输出数据格式

scikit-learn Transformers

Feature extraction

• tsfresh.FeatureAugmenter

Feature selection

• tsfresh.FeatureSelector

Feature extraction and selection

• tsfresh.RelevantFeatureAugmenter

from sklearn.pipeline import Pipeline
from sklearn.ensemble import RandomForestClassifier
from tsfresh.examples import load_robot_execution_failures
from tsfresh.transformers import RelevantFeatureAugmenter
import pandas as pd

# download data
from tsfresh.examples.robot_execution_failures import download_robot_execution_failures
download_robot_execution_failures()

pipeline = Pipeline([
    ("augmenter", RelevantFeatureAugmenter(column_id = "id", column_sort = "time")),
    ("classifier", RandomForestClassifier()),
])

df_ts, y = load_robot_execution_failures()
X = pd.DataFrame(index = y.index)

pipeline.set_params(augmenter__timeseries_container = df_ts)
pipeline.fit(X, y)

大数据

Dask

PySpark

Rolling 和 时间序列预测

参考

• tsfresh Github
• tsfresh 文档
• tsfresh 使用小结