https://github.com/sktime/pytorch-forecasting

Time series forecasting with PyTorch

Science Score: 46.0%

This score indicates how likely this project is to be science-related based on various indicators:

○
CITATION.cff file
✓
codemeta.json file
Found codemeta.json file
✓
.zenodo.json file
Found .zenodo.json file
○
DOI references
✓
Academic publication links
Links to: arxiv.org, sciencedirect.com
✓
Committers with academic emails
3 of 80 committers (3.8%) from academic institutions
○
Institutional organization owner
○
JOSS paper metadata
○
Scientific vocabulary similarity
Low similarity (14.1%) to scientific vocabulary

Keywords

ai artifical-intelligense data-science deep-learning forecasting gpu machine-learning neural-networks pandas python pytorch pytorch-lightning temporal timeseries timeseries-forecasting uncertainty

Keywords from Contributors

data-mining distributed jax large-language-models sktime time-series-classification time-series-regression time-series-segmentation anomaly-detection changepoint-detection

Last synced: 5 months ago · JSON representation

Repository

Time series forecasting with PyTorch

Basic Info

Host: GitHub
Owner: sktime
License: mit
Language: Python
Default Branch: main
Homepage: https://pytorch-forecasting.readthedocs.io/
Size: 40.7 MB

Statistics

Stars: 4,468
Watchers: 43
Forks: 699
Open Issues: 585
Releases: 36

Topics

ai artifical-intelligense data-science deep-learning forecasting gpu machine-learning neural-networks pandas python pytorch pytorch-lightning temporal timeseries timeseries-forecasting uncertainty

Created over 5 years ago · Last pushed 6 months ago

Metadata Files

Readme Changelog License Codeowners

PyTorch Forecasting is a PyTorch-based package for forecasting with state-of-the-art deep learning architectures. It provides a high-level API and uses PyTorch Lightning to scale training on GPU or CPU, with automatic logging.

| | Documentation · Tutorials · Release Notes | |---|---| | Open Source | | | | Community | | | CI/CD | | | Code | | | Downloads | PyPI - Downloads |

Our article on Towards Data Science introduces the package and provides background information.

PyTorch Forecasting aims to ease state-of-the-art timeseries forecasting with neural networks for real-world cases and research alike. The goal is to provide a high-level API with maximum flexibility for professionals and reasonable defaults for beginners. Specifically, the package provides

A timeseries dataset class which abstracts handling variable transformations, missing values, randomized subsampling, multiple history lengths, etc.
A base model class which provides basic training of timeseries models along with logging in tensorboard and generic visualizations such actual vs predictions and dependency plots
Multiple neural network architectures for timeseries forecasting that have been enhanced for real-world deployment and come with in-built interpretation capabilities
Multi-horizon timeseries metrics
Hyperparameter tuning with optuna

The package is built on pytorch-lightning to allow training on CPUs, single and multiple GPUs out-of-the-box.

Installation

If you are working on windows, you need to first install PyTorch with

pip install torch -f https://download.pytorch.org/whl/torch_stable.html.

Otherwise, you can proceed with

pip install pytorch-forecasting

Alternatively, you can install the package via conda

conda install pytorch-forecasting pytorch -c pytorch>=1.7 -c conda-forge

PyTorch Forecasting is now installed from the conda-forge channel while PyTorch is install from the pytorch channel.

To use the MQF2 loss (multivariate quantile loss), also install pip install pytorch-forecasting[mqf2]

Documentation

Visit https://pytorch-forecasting.readthedocs.io to read the documentation with detailed tutorials.

Available models

The documentation provides a comparison of available models.

Temporal Fusion Transformers for Interpretable Multi-horizon Time Series Forecasting which outperforms DeepAR by Amazon by 36-69% in benchmarks
N-BEATS: Neural basis expansion analysis for interpretable time series forecasting which has (if used as ensemble) outperformed all other methods including ensembles of traditional statical methods in the M4 competition. The M4 competition is arguably the most important benchmark for univariate time series forecasting.
N-HiTS: Neural Hierarchical Interpolation for Time Series Forecasting which supports covariates and has consistently beaten N-BEATS. It is also particularly well-suited for long-horizon forecasting.
DeepAR: Probabilistic forecasting with autoregressive recurrent networks which is the one of the most popular forecasting algorithms and is often used as a baseline
Simple standard networks for baselining: LSTM and GRU networks as well as a MLP on the decoder
A baseline model that always predicts the latest known value

To implement new models or other custom components, see the How to implement new models tutorial. It covers basic as well as advanced architectures.

Usage example

Networks can be trained with the PyTorch Lighning Trainer on pandas Dataframes which are first converted to a TimeSeriesDataSet.

```python

imports for training

import lightning.pytorch as pl from lightning.pytorch.loggers import TensorBoardLogger from lightning.pytorch.callbacks import EarlyStopping, LearningRateMonitor

import dataset, network to train and metric to optimize

from pytorch_forecasting import TimeSeriesDataSet, TemporalFusionTransformer, QuantileLoss from lightning.pytorch.tuner import Tuner

load data: this is pandas dataframe with at least a column for

* the target (what you want to predict)

* the timeseries ID (which should be a unique string to identify each timeseries)

* the time of the observation (which should be a monotonically increasing integer)

data = ...

define the dataset, i.e. add metadata to pandas dataframe for the model to understand it

maxencoderlength = 36 maxpredictionlength = 6 training_cutoff = "YYYY-MM-DD" # day for cutoff

training = TimeSeriesDataSet( data[lambda x: x.date <= trainingcutoff], timeidx= ..., # column name of time of observation target= ..., # column name of target to predict groupids=[ ... ], # column name(s) for timeseries IDs maxencoderlength=maxencoderlength, # how much history to use maxpredictionlength=maxpredictionlength, # how far to predict into future # covariates static for a timeseries ID staticcategoricals=[ ... ], staticreals=[ ... ], # covariates known and unknown in the future to inform prediction timevaryingknowncategoricals=[ ... ], timevaryingknownreals=[ ... ], timevaryingunknowncategoricals=[ ... ], timevaryingunknown_reals=[ ... ], )

create validation dataset using the same normalization techniques as for the training dataset

validation = TimeSeriesDataSet.fromdataset(training, data, minpredictionidx=training.index.time.max() + 1, stoprandomization=True)

convert datasets to dataloaders for training

batchsize = 128 traindataloader = training.todataloader(train=True, batchsize=batchsize, numworkers=2) valdataloader = validation.todataloader(train=False, batchsize=batchsize, num_workers=2)

create PyTorch Lighning Trainer with early stopping

earlystopcallback = EarlyStopping(monitor="valloss", mindelta=1e-4, patience=1, verbose=False, mode="min") lrlogger = LearningRateMonitor() trainer = pl.Trainer( maxepochs=100, accelerator="auto", # run on CPU, if on multiple GPUs, use strategy="ddp" gradientclipval=0.1, limittrainbatches=30, # 30 batches per epoch callbacks=[lrlogger, earlystopcallback], logger=TensorBoardLogger("lightninglogs") )

define network to train - the architecture is mostly inferred from the dataset, so that only a few hyperparameters have to be set by the user

tft = TemporalFusionTransformer.fromdataset( # dataset training, # architecture hyperparameters hiddensize=32, attentionheadsize=1, dropout=0.1, hiddencontinuoussize=16, # loss metric to optimize loss=QuantileLoss(), # logging frequency loginterval=2, # optimizer parameters learningrate=0.03, reduceonplateau_patience=4 ) print(f"Number of parameters in network: {tft.size()/1e3:.1f}k")

find the optimal learning rate

res = Tuner(trainer).lrfind( tft, traindataloaders=traindataloader, valdataloaders=valdataloader, earlystopthreshold=1000.0, maxlr=0.3, )

and plot the result - always visually confirm that the suggested learning rate makes sense

print(f"suggested learning rate: {res.suggestion()}") fig = res.plot(show=True, suggest=True) fig.show()

fit the model on the data - redefine the model with the correct learning rate if necessary

trainer.fit( tft, traindataloaders=traindataloader, valdataloaders=valdataloader, ) ```

Owner

Name: sktime
Login: sktime
Kind: organization
Email: sktime.toolbox@gmail.com

Website: https://www.sktime.net
Repositories: 29
Profile: https://github.com/sktime

A unified framework for machine learning with time series

Committers

Last synced: 6 months ago

All Time

Total Commits: 1,368
Total Committers: 80
Avg Commits per committer: 17.1
Development Distribution Score (DDS): 0.576

Past Year

Commits: 130
Committers: 35
Avg Commits per committer: 3.714
Development Distribution Score (DDS): 0.669

Top Committers

Name	Email	Commits
Jan Beitner	b**n@b**m	580
dependabot[bot]	4**]@u**m	354
dependabot-preview[bot]	2**]@u**m	115
Franz Király	f**y@u**k	54
Franz Király	f**y@g**i	43
dehoyosb	d**2@g**m	27
Jan Beitner	j**r@i**m	25
Felix Hirwa Nshuti	h**x@g**m	15
Jake F	j**f@b**m	13
Aryan Saini	1**x@u**m	10
Pranav Bhat	6**P@u**m	10
Xinyu-Wu-0000	x**u@o**m	8
Daniil Gafni	d**i@i**u	7
Jan Beitner	j**r@g**m	5
Jirka Borovec	6**a@u**m	4
Li Yu	l**8@g**m	4
Luke Merrick	l**k@m**i	4
Xinyu Wu	5**u@u**m	4
Ben Steel	b**l@g**m	3
Jirka	j**c@s**z	3
Julio Antonio Soto	j**o@e**s	3
sirius	a**s@g**m	3
Anirban Ray	3**a@u**m	2
Christy Bergman	c**n@g**m	2
Elias Nehme	e**2@g**m	2
Ewan Thompson	e**t@g**m	2
JustinNeumann	J**n@u**m	2
NISHANT KUMAR	a**7@g**m	2
Oleg	p**o@g**m	2
Timo Klerx	t**k@m**e	2
and 50 more...

Committer Domains (Top 20 + Academic)

qq.com: 2 mailbox.org: 1 szepe.net: 1 krounix.org: 1 luckydinosaur.com: 1 outlook.it: 1 wp.pl: 1 naver.com: 1 histark.com: 1 student-cs.fr: 1 boun.edu.tr: 1 rozetka.ua: 1 sumup.com: 1 iti.es: 1 cngmi.com: 1 mercadolibre.com.co: 1 unal.edu.co: 1 mail.upb.de: 1 esbet.es: 1 seznam.cz: 1 ucl.ac.uk: 1

Issues and Pull Requests

Last synced: 5 months ago

All Time

Total issues: 166
Total pull requests: 280
Average time to close issues: 9 months
Average time to close pull requests: about 1 month
Total issue authors: 85
Total pull request authors: 40
Average comments per issue: 1.81
Average comments per pull request: 3.45
Merged pull requests: 170
Bot issues: 0
Bot pull requests: 21

Past Year

Issues: 134
Pull requests: 225
Average time to close issues: 22 days
Average time to close pull requests: 14 days
Issue authors: 53
Pull request authors: 24
Average comments per issue: 0.9
Average comments per pull request: 3.68
Merged pull requests: 136
Bot issues: 0
Bot pull requests: 16

View more stats

Top Authors

Issue Authors

fkiraly (31)
PranavBhatP (18)
phoeenniixx (9)
jobs-git (8)
cngmid (5)
arizzuto (3)
Neronjust2017 (3)
workhours (3)
moghadas76 (3)
benHeid (3)
hagersalehahmed (2)
mrgreen3325 (2)
JeremyKulcsarDS (2)
mkuiack (2)
grudloff (2)

Pull Request Authors

fkiraly (111)
phoeenniixx (28)
PranavBhatP (24)
dependabot[bot] (21)
fnhirwa (20)
Borda (7)
RUPESH-KUMAR01 (6)
jobs-git (4)
Vishnu-Rangiah (4)
cngmid (4)
julian-fong (4)
XinyuWuu (4)
yarnabrina (3)
ewth (3)
jdb78 (3)

Top Labels

Issue Labels

bug (56) enhancement (46) documentation (13) maintenance (8) API design (6) module:models (4) question (3) feature request (3) good first issue (3) module:test_framework (2) dependencies (1) new network (1) module:metrics (1)

Pull Request Labels

maintenance (95) enhancement (45) documentation (22) dependencies (18) bug (18) release (13) module:models (7) API design (6) module:test_framework (6) new network (4) module:metrics (2)

Packages

Total packages: 2
Total downloads:
- pypi 259,879 last-month
Total docker downloads: 127

Total dependent packages: 9
(may contain duplicates)
Total dependent repositories: 34
(may contain duplicates)
Total versions: 75
Total maintainers: 2

pypi.org: pytorch-forecasting

Forecasting timeseries with PyTorch - dataloaders, normalizers, metrics and models

Documentation: https://pytorch-forecasting.readthedocs.io/
License: MIT License
Latest release: 1.4.0
published 8 months ago

Versions: 39
Dependent Packages: 9
Dependent Repositories: 34
Downloads: 259,879 Last month
Docker Downloads: 127

Rankings

Dependent packages count: 0.8%

Downloads: 1.2%

Stargazers count: 1.3%

Average: 1.8%

Forks count: 2.3%

Dependent repos count: 2.5%

Docker downloads count: 2.8%

Maintainers (2)

jdb78 fkiraly

Last synced: 6 months ago

proxy.golang.org: github.com/sktime/pytorch-forecasting

Documentation: https://pkg.go.dev/github.com/sktime/pytorch-forecasting#section-documentation
License: mit
Latest release: v1.4.0
published 8 months ago

Versions: 36
Dependent Packages: 0
Dependent Repositories: 0

Rankings

Dependent packages count: 5.5%

Average: 5.6%

Dependent repos count: 5.8%

Last synced: 6 months ago

Dependencies

.github/workflows/autoapprove.yml actions

dependabot/fetch-metadata v1.1.1 composite

.github/workflows/automerge.yml actions

dependabot/fetch-metadata v1.1.1 composite

.github/workflows/lint.yml actions

actions/checkout v2 composite
actions/setup-python v2 composite

.github/workflows/pypi_release.yml actions

actions/checkout v2 composite
actions/setup-python v2 composite

.github/workflows/test.yml actions

actions/cache v2 composite
actions/checkout v2 composite
actions/setup-python v2 composite
actions/setup-python v1 composite
actions/upload-artifact v2 composite
codecov/codecov-action v1 composite

docs/requirements.txt pypi

cloudpickle *
docutils *
fastapi >0.80
ipython *
lightning >=2.0.0
matplotlib *
nbconvert >=6.3.0
nbsphinx *
optuna >=3.1.0
pandas >=1.3
pandoc *
pydata-sphinx-theme *
pytorch-optimizer >=2.5.1
recommonmark >=0.7.1
scikit-learn >1.2
scipy *
sphinx >3.2
statsmodels *
torch >=2.0,

poetry.lock pypi

215 dependencies

pyproject.toml pypi

cpflows ^0.1.2
fastapi >=0.80
lightning ^2.0.0
matplotlib *
networkx ^3.0.0
optuna ^3.1.0
pandas >=1.3.0,<=3.0.0
pytest-github-actions-annotate-failures *
python >=3.8,<3.11
pytorch-optimizer ^2.5.1
scikit-learn ^1.2
scipy ^1.8
statsmodels *
torch ^2.0.0,!=2.0.1

https://github.com/sktime/pytorch-forecasting

Science Score: 46.0%

Keywords

Keywords from Contributors

Repository

Basic Info

Statistics

Topics

Metadata Files

README.md

Installation

Documentation

Available models

Usage example

imports for training

import dataset, network to train and metric to optimize

load data: this is pandas dataframe with at least a column for

* the target (what you want to predict)

* the timeseries ID (which should be a unique string to identify each timeseries)

* the time of the observation (which should be a monotonically increasing integer)

define the dataset, i.e. add metadata to pandas dataframe for the model to understand it

create validation dataset using the same normalization techniques as for the training dataset

convert datasets to dataloaders for training

create PyTorch Lighning Trainer with early stopping

define network to train - the architecture is mostly inferred from the dataset, so that only a few hyperparameters have to be set by the user

find the optimal learning rate

and plot the result - always visually confirm that the suggested learning rate makes sense

fit the model on the data - redefine the model with the correct learning rate if necessary

Owner

Committers

All Time

Past Year

Top Committers

Committer Domains (Top 20 + Academic)

Issues and Pull Requests

All Time

Past Year

Top Authors

Issue Authors

Pull Request Authors

Top Labels

Issue Labels

Pull Request Labels

Packages

pypi.org: pytorch-forecasting

Rankings

Maintainers (2)

proxy.golang.org: github.com/sktime/pytorch-forecasting

Rankings

Dependencies