Time Series Forecasting
Last updated
Was this helpful?
Last updated
Was this helpful?
Univariate LSTM Models : one observation time-series data, predict the next value in the sequence
Multivariate LSTM Models : two or more observation time-series data, predict the next value in the sequence
Multiple Input Series : two or more parallel input time series and an output time series that is dependent on the input time series
Multiple Parallel Series : multiple parallel time series and a value must be predicted for each
Univariate Multi-Step LSTM Models : one observation time-series data, predict the multi step value in the sequence prediction.
Multivariate Multi-Step LSTM Models : two or more observation time-series data, predict the multi step value in the sequence prediction.
Multiple Input Multi-Step Output.
Multiple Parallel Input and Multi-Step Output.
, ,
, ,
: MPI-MSO
: LSTM
: ANFIS
Dataset : Electricity Load ENTSO, Model : LSTM, Type: Multivariate
, Dataset: river flow , flood severity, Model: LSTM, Transformer, Simple Multi-Head Attention, Transformer with a linear decoder, DA-RNN, Transformer XL, Informer, DeepAR
import statsmodels
import keras
Nowcasting, Upsampling, Interpolation, Super resolution
top open source deep learning for time series forecasting frameworks.
XGboost, LGBM, pmdarima, stanpy (for bayesian modelling)
Prophet - seems to be the current 'standard' choice
ARIMA - Classical choice
Exponential Moving Average - dead simple to implement, works well for stuff that's a time series but not very seasonal
Kalman/Statespace model - used by Splunk's predict[1] command (pretty sure I always used LLP5)
Prophet, statsmodels, tf.keras for RNNs.
tensorflow probability's time series package.
PyTorch for recurrent nets
State of the art is 1D convnets, bleeding edge is transformers.
pycaret timeseries
lgbm light gbm
cvxpy
TensorFlow's LSTMCell
sktime
bssts
statsmodels
: VAR
: VAR
: VAR
: VAR
: VAR
: NN
: LSTM
Dataset: Sunspot Zurich, Model: LSTM
Dataset: Air Pollution, Model: LSTM
Dataset: Stock Market, Model: LSTM, RF
Dataset: Amazon Stock Model: LSTM with addition
Dataset: Electric Consumption Model: LSTM
Dataset: Electricity Consumption Model: LSTM
Dataset: ENSO El Nino, Model: LSTM
Dataset: Sensor, Model: LSTM
Dataset: Solar power, Model: LSTM
Dataset: English power consumption, Model: ANP-RNN "Recurrent Attentive Neural Process for Sequential Data", ANP: Attentive Neural Processes, NP: Neural Processes, LSTM
, Dataset: Kaggle energy consuption, Model: LSTM, Seq2Seq
: LSTM, ARIMA, Prophet
Dataset: Electricity, Model: Feed forward Neural Network FFNN, Simple Moving Average SMA, Weighted Moving Average WMA, Simple Exponential Smoothing SES, Holts Winters HW, Autoregressive Integrated Moving Average ARIMA, Recurrent Neural Networks RNN, Long Short Term Memory cells LSTM, Gated Recurrent Unit cells GRU, Type: Univariate
Dataset: Daily Temperature, Model: LSTM
Dataset: Pollution, Solar Energy, Traffic data etec. Model MLP,RNN,LSTM,GRU, ARIMA, SVR, RF and TSR-RNN
Dataset: ?? Model: ARIMA, VAR, LSTM
Dataset: Olypic, LeBron, Zika, Model: ARIMA dan LSTM
Dataset: Power usage, Model: linear regression, random forest, support vector machine, BP neural network, GRU, LSTM
Dataset: precipitation, Model: ARIMA, LSTM, FNN(Feed forward Neural Networks), TLNN(Time lagged Neural Networks), SANN(Seasonal Artificial Neural Networks
(Note: VAR should only for )
: VAR
: VAR
: VAR
: VAR
: VAR
,
Collection of papers
One stop shop for time series analysis in Python
This framework by Amazon remains one of the top DL based time series forecasting frameworks on GitHub. However, there are some down sides including lock-in to MXNet (a rather obscure architecture). The repository also doesn’t seem to be quick at adding new research.
This is an upcoming PyTorch based deep learning for time series forecasting framework. The repository features a lot of recent models out of research conferences along with an easy to use deployment API. The repository is one of the few repos to have new models, coverage tests, and interpretability metrics.
This is another time series forecasting repository. Unfortunately it looks like particularly recent activity has diminished on it.
Another framework, written in PyTorch, this repository focuses more on probabilistic models. The repository isn’t that active (last commit was in November).
engineering statistics
Prophet
in R
in Python
in R
LSTMs have been going the way of the dinosaurs since 2018. If you really need a complex neural network (over 1D convolution approaches), transformers are the current SOTA.
# Statistical vs Deep Learning forecasting methods