时序分析Py工具包_from pyts.forecasting importoltwinters-CSDN博客

本文链接：https://blog.csdn.net/O3OTZ/article/details/134305388

prophet

github ⭐17.1k

仅适用单变量的时序预测模型
仅支持 prophet 这一个模型。

tsfresh

github ⭐7.8k

用于时序特征工程，包括对时间序列自动提取特征、特征选择、特征转换等
不能用于实现时序预测或时序分类等任务。

sktime

github ⭐7k

具有时序预测、时序分类、时序回归、时序聚类等
可与 scikit-learn 进行互操作
提供经典的统计学模型系列（ARIMA、ETS、prophet等）
提供模型Ensemble能力和AutoML功能（带有模型选择和自动调参功能的时序建模）
也提供一些深度学习的模型，如Transformer等

功能表

Module	Status	Links
[Forecasting]	stable	Tutorial · API Reference · Extension Template
[Time Series Classification]	stable	Tutorial · API Reference · Extension Template
[Time Series Regression]	stable	API Reference
[Transformations]	stable	Tutorial · API Reference · Extension Template
[Parameter fitting]	maturing	API Reference · Extension Template
[Time Series Clustering]	maturing	API Reference · Extension Template
[Time Series Distances/Kernels]	maturing	Tutorial · API Reference · Extension Template
[Time Series Alignment]	experimental	API Reference · Extension Template
[Annotation]	experimental	Extension Template
[Distributions and simulation]	experimental

darts

github ⭐6.6k

时序分析工具的集大成者，支持单变量和多变量预测
聚焦于时序预测问题，支持从ARIMA到深度学习模型，如Transformer、TCN等新的时序建模方法。
该库还可以方便地对模型进行回溯测试，并将多个模型的预测和外部回归组合起来。

可用模型

Model	Sources	Target Series Support: Univariate/ Multivariate	Covariates Support: Past-observed/ Future-known/ Static	Probabilistic Forecasting: Sampled/ Distribution Parameters	Training & Forecasting on Multiple Series
Baseline Models (LocalForecastingModel)
NaiveMean		🟩 🟥	🟥 🟥 🟥	🟥 🟥	🟥
NaiveSeasonal		🟩 🟥	🟥 🟥 🟥	🟥 🟥	🟥
NaiveDrift		🟩 🟥	🟥 🟥 🟥	🟥 🟥	🟥
NaiveMovingAverage		🟩 🟥	🟥 🟥 🟥	🟥 🟥	🟥
Statistical / Classic Models (LocalForecastingModel)
ARIMA		🟩 🟥	🟥 🟩 🟥	🟩 🟥	🟥
VARIMA		🟥 🟩	🟥 🟩 🟥	🟥 🟥	🟥
AutoARIMA		🟩 🟥	🟥 🟩 🟥	🟥 🟥	🟥
StatsForecastAutoArima (faster AutoARIMA)	Nixtla’s statsforecast	🟩 🟥	🟥 🟩 🟥	🟩 🟥	🟥
ExponentialSmoothing		🟩 🟥	🟥 🟥 🟥	🟩 🟥	🟥
StatsforecastAutoETS	Nixtla’s statsforecast	🟩 🟥	🟥 🟩 🟥	🟩 🟥	🟥
StatsforecastAutoCES	Nixtla’s statsforecast	🟩 🟥	🟥 🟥 🟥	🟥 🟥	🟥
BATS and TBATS	TBATS paper	🟩 🟥	🟥 🟥 🟥	🟩 🟥	🟥
Theta and FourTheta	Theta & 4 Theta	🟩 🟥	🟥 🟥 🟥	🟥 🟥	🟥
StatsForecastAutoTheta	Nixtla’s statsforecast	🟩 🟥	🟥 🟥 🟥	🟩 🟥	🟥
Prophet (see install notes)	Prophet repo	🟩 🟥	🟥 🟩 🟥	🟩 🟥	🟥
FFT (Fast Fourier Transform)		🟩 🟥	🟥 🟥 🟥	🟥 🟥	🟥
KalmanForecaster using the Kalman filter and N4SID for system identification	N4SID paper	🟩 🟩	🟥 🟩 🟥	🟩 🟥	🟥
Croston method		🟩 🟥	🟥 🟥 🟥	🟥 🟥	🟥
Regression Models (GlobalForecastingModel)
RegressionModel: generic wrapper around any sklearn regression model		🟩 🟩	🟩 🟩 🟩	🟥 🟥	🟩
LinearRegressionModel		🟩 🟩	🟩 🟩 🟩	🟩 🟩	🟩
RandomForest		🟩 🟩	🟩 🟩 🟩	🟥 🟥	🟩
LightGBMModel,		🟩 🟩	🟩 🟩 🟩	🟩 🟩	🟩
XGBModel		🟩 🟩	🟩 🟩 🟩	🟩 🟩	🟩
CatBoostModel		🟩 🟩	🟩 🟩 🟩	🟩 🟩	🟩
PyTorch (Lightning)-based Models (GlobalForecastingModel)
RNNModel (incl. LSTM and GRU); equivalent to DeepAR in its probabilistic version	DeepAR paper	🟩 🟩	🟥 🟩 🟥	🟩 🟩	🟩
BlockRNNModel (incl. LSTM and GRU)		🟩 🟩	🟩 🟥 🟥	🟩 🟩	🟩
NBEATSModel	N-BEATS paper	🟩 🟩	🟩 🟥 🟥	🟩 🟩	🟩
NHiTSModel	N-HiTS paper	🟩 🟩	🟩 🟥 🟥	🟩 🟩	🟩
TCNModel	TCN paper, DeepTCN paper, blog post	🟩 🟩	🟩 🟥 🟥	🟩 🟩	🟩
TransformerModel		🟩 🟩	🟩 🟥 🟥	🟩 🟩	🟩
TFTModel (Temporal Fusion Transformer)	TFT paper, PyTorch Forecasting	🟩 🟩	🟩 🟩 🟩	🟩 🟩	🟩
DLinearModel	DLinear paper	🟩 🟩	🟩 🟩 🟩	🟩 🟩	🟩
NLinearModel	NLinear paper	🟩 🟩	🟩 🟩 🟩	🟩 🟩	🟩
TiDEModel	TiDE paper	🟩 🟩	🟩 🟩 🟩	🟩 🟩	🟩
Ensemble Models (GlobalForecastingModel): Model support is dependent on ensembled forecasting models and the ensemble model itself
NaiveEnsembleModel		🟩 🟩	🟩 🟩 🟩	🟩 🟩	🟩
RegressionEnsembleModel		🟩 🟩	🟩 🟩 🟩	🟩 🟩	🟩

Kats

github ⭐4.6k

使用此库，可以执行以下操作：

时间序列分析
模式检测，包括季节性、异常值、趋势变化
产生65个特征的特征工程模块
对时间序列数据建立预测模型，包括Prophet、ARIMA、Holt Winters等。

gluonts

github ⭐3.9k

可用模型

Model + Paper	Local/global	Data layout	Architecture/method	Implementation
DeepAR Salinas et al. 2020	Global	Univariate	RNN	MXNet, PyTorch
DeepState Rangapuram et al. 2018	Global	Univariate	RNN, state-space model	MXNet
DeepFactor Wang et al. 2019	Global	Univariate	RNN, state-space model, Gaussian process	MXNet
Deep Renewal Processes Türkmen et al. 2021	Global	Univariate	RNN	MXNet
GPForecaster	Global	Univariate	MLP, Gaussian process	MXNet
MQ-CNN Wen et al. 2017	Global	Univariate	CNN encoder, MLP decoder	MXNet
MQ-RNN Wen et al. 2017	Global	Univariate	RNN encoder, MLP encoder	MXNet
N-BEATS Oreshkin et al. 2019	Global	Univariate	MLP, residual links	MXNet
Rotbaum Hasson et al. 2021	Global	Univariate	XGBoost, Quantile Regression Forests, LightGBM, Level Set Forecaster	Numpy
Temporal Fusion Transformer Lim et al. 2021	Global	Univariate	LSTM, self attention	MXNet, PyTorch
Transformer Vaswani et al. 2017	Global	Univariate	MLP, multi-head attention	MXNet
WaveNet van den Oord et al. 2016	Global	Univariate	Dilated convolution	MXNet, PyTorch
SimpleFeedForward	Global	Univariate	MLP	MXNet, PyTorch
DeepNPTS	Global	Univariate	MLP	PyTorch
MQF2 Kan et al. 2022	Global	Univariate	RNN, ICNN	PyTorch
DeepVAR Salinas et al. 2019	Global	Multivariate	RNN	MXNet
GPVAR Salinas et al. 2019	Global	Multivariate	RNN, Gaussian process	MXNet
LSTNet Lai et al. 2018	Global	Multivariate	LSTM	MXNet
DeepTPP Shchur et al. 2020	Global	Multivariate events	RNN, temporal point process	MXNet
DeepVARHierarchical Rangapuram et al. 2021	Global	Hierarchical	RNN	MXNet
RForecast Hyndman et al. 2008	Local	Univariate	ARIMA, ETS, Croston, TBATS	Wrapped R package
Prophet Taylor et al. 2017	Local	Univariate	-	Wrapped Python package
NaiveSeasonal Hyndman et al. 2018	Local	Univariate	-	Numpy
Naive2 Makridakis et al. 1998	Local	Univariate	-	Numpy
NPTS	Local	Univariate	-	Numpy

Merlion

github ⭐3.1k

支持时序预测（Forecasting）和异常检测（Anomaly Detection）
支持单变量和多变量的时序分析
支持了模型融合（Ensemble）以及AutoML能力（带有模型选择和自动调参功能的时序建模）
对于时序预测任务，支持统计学模型和机器学习模型，其中统计学模型包括ARIMA、ETS、Prophet等；而机器学习模型则主要是基于决策树的集成模型，例如RF和GB等。
支持自动绘制真实值和预测结果及置信区间的对比曲线

tslearn

github ⭐2.6k

针对时序数据（time series）的机器学习相关工具包
可与 scikit-learn 进行互操作
具有数据预处理、分类、回归（预测是一种特殊形式的回归任务）以及聚类等功能
tslearn可以与其他时序工具包进行整合使用，如scikit-learn、tsfresh、sktime及pyts等

功能表

data	processing	clustering	classification	regression	metrics
UCR Datasets	Scaling	TimeSeriesKMeans	KNN Classifier	KNN Regressor	Dynamic Time Warping
Generators	Piecewise	KShape	TimeSeriesSVC	TimeSeriesSVR	Global Alignment Kernel
Conversion(1, 2)		KernelKmeans	LearningShapelets	MLP	Barycenters
			Early Classification		Matrix Profile