This note lists Python libraries relevant to time series prediction. They are ranked by monthly downloads in the last 30 days, which is no guarantee of quality. For some we've added a "hello world" example in timeseries-notebooks, to help you cut through the many different conventions. A growing number are exposed in a common functional signature explained in the timemachines package. During testing those are assigned elo ratings based on performance out of sample on live data
Pull requests at timeseries-notebooks are most welcome.
| Downloads | Hello? | Rated? | Remarks, examples | |
| pykalman | 81,857 | notebook | ||
| filterpy | 25,004 | notebook | no estimation | |
| control | 22,053 | |||
| pydlm | 18,766 | notebook | elo | |
| simdkalman | 8,173 | notebook | vectorized | |
| river | 6,581 | notebook | formerly "creme" | |
| timemachines | 6,405 | notebooks | elo | creates Elo ratings |
| timecast | 399 | |||
| proust | 341 | |||
| onelearn | 330 | |||
| state-space | 91 | symbolic derivs | ||
| The following are not packaged | ||||
| robust-kalman | github paper |
Got more ideas? Please file an issue. You might also be interested in this review paper.
| Downloads | Hello? | Rated? | Remarks, examples | |
| fbprophet | 769,124 | notebook | elo | reviewed discussion |
| pmdarima | 300,638 | notebook | ||
| pygam | 175,589 | |||
| hmmlearn | 122,000 | |||
| autogluon | 71,908 | |||
| stldecompose | 65,607 | |||
| gluonts | 23,459 | paper paper | ||
| cronston | 22,459 | |||
| forecast | 22,207 | broken | ||
| tbats | 17,200 | |||
| timeseries | 14,811 | |||
| pyflux | 12,773 | notebook | ||
| ppscore | 5,596 | |||
| pyculiar | 5,031 | |||
| hcrystalball | 4,500 | |||
| cesium | 4,422 | |||
| neuralprophet | 4,200 | notebook | discussion | |
| pytorch-forecasting | 3,372 | notebook | notebook almost done | |
| seglearn | 3,100 | |||
| auto_ts | 2794 | notebook | ||
| autots | 2,294 | |||
| tsai | 2,635 | |||
| skits | 2,630 | |||
| nitime | 2,441 | |||
| time-series-predictor | 2,021 | |||
| nbeats-keras | 1,898 | paper post c.f. nbeats-forecast, nbeats etc | ||
| pytorchts | 1,759 | paper | ||
| pydbm | 1,700 | |||
| greykite | 1,512 | paper | ||
| da-dapper | 1,501 | |||
| autotime | 1,360 | |||
| orbit-ml | 1,400 | article docs Uber | ||
| u8darts | 1,355 | article | ||
| atspy | 1,125 | |||
| tsmoothie | 1,012 | |||
| causality | 1,008 | broken | ||
| autotsf | 1,000 | |||
| flood-forecast | 1,000 | a.k.a. flow-forecast | ||
| anticipy | 934 | |||
| arimafd | 898 | |||
| pure-predict | 801 | lambda deploy | ||
| pyCWT | 739 | |||
| pyinform | 744 | |||
| gendis | 692 | |||
| scikit-hts | 644 | M5 | ||
| cond-rnn | 640 | |||
| scikit-hts | 632 | hierarchical | ||
| thyus-timeseries | 557 | |||
| scifin | 573 | |||
| openseries | 538 | |||
| autoregressive | 488 | |||
| va-timeseries | 465 | |||
| pyaf | 460 | |||
| theano-lstm | 430 | |||
| esrnn | 417 | paper | ||
| firets | 390 | |||
| darts | 380 | |||
| time-series-generator | 375 | keras | ||
| proust | 341 | online | ||
| tods | 323 | |||
| pyfts | 298 | fuzzy | ||
| pysf | 269 | |||
| timeseries_fastai | 274 | |||
| giotto-time | 266 | |||
| pyFTS | 263 | |||
| pandas-timeseries | 214 | |||
| pycatchmod | 269 | |||
| timecopts | 199 | |||
| fecon236 | 187 | |||
| phased_lstm_keras | 177 | |||
| u8timeseries | 155 | |||
| tendency-predict | 150 | |||
| rnn | 109 | |||
| tsf | 103 | |||
| pytisean | 98 | wraps TISEAN | ||
| karas-on-lstm | 90 | |||
| autoregression | 80 | |||
| raptorarima | 60 | |||
| pyda | 59 | |||
| divinity | 54 | notebook | ||
| The following are not packaged on PyPI | ||||
| informer | AAAI '21 best paper | |||
| fforma | M5 discussion | |||
| dcrnn | github paper | |||
| tensor-learning | github paper multivariate, traffic | |||
| lstnet | github paper lstm traffic | |||
| dpsom | github paper clustering | |||
| dilate | github paper | |||
| mtgnn | github paper graph NN | |||
| dsanet | github paper transformer | |||
| dts | github paper electricity | |||
| deepglow | github paper | |||
| dvmst-net | github paper | |||
| forecastnet | github paper | |||
| mlcnn-multivariate-time-series | github paper | |||
| mlip_m5 | github paper | |||
| forgan | github paper | |||
| gru_ode_bayes | github paper sporadic | |||
| latent_ode | github paper irregular | |||
| processsequenceprediction | github paper | |||
| cycle_prediction | github doc arrival | |||
| seriesnet | github paper |
If I have left something obvious out, please add it to the timeseries by popularity column. Download statistics can be misleading, of course, and I think we all know the better metric is how well they perform in real-time on live data. Don't forget that many algorithms on the leaderboards have code badges allowing you to click through to the implementation.
General purpose analytic libraries such as scipy and statsmodels are not included, though it goes without saying the time series capabilities are popular. You'd be crazy not to consider the big ones, though here the download statistics are even less relevant as time series is only a small part of their capability. Other lists include paperswithcode and awesome_time_series_in_python
| Monthly downloads | Notes | |
| tensorflow | 4,586,049 | hmm etc |
| sklearn | 2,157,986 | hmm -> hmmlearn, listed above |
| numpy | 42,091,102 | |
| scipy | 14,107,552 | scipy.stats etc |
| statsmodels | 2,552,523 | statsmodels.tsa |
| pandas | 26,263,006 |
New category. Stay tuned.
| Downloads | Hello ? | Remarks, examples | ||
| tsai | 3,842 | paper, paper, paper and more see list in README | ||
| seglearn | 1,823 | |||
| time-series-generator | 800 | hack for keras multi-step outputs | ||
| seq2seq | 204 | |||
| The following are not on PyPI | ||||
| LSTM-Neural-Network-for-Time-Series-Prediction | ||||
| DeepTimeSeries | ||||
| ECR-LSTM | ||||
| Deep-Time-Series-Prediction | ||||
| seq2seq-signal-prediction | ||||
| DualAttentionSeq2Seq | ||||
| TimeSeries-Seq2Seq-deepLSTMs-Keras | ||||
| TCN_Seq2Seq | ||||
| tcn-keras-Examples | examples | |||
| DynamicAttentionNetworks | electricity | |||
| TimeSeries_Seq2Seq | notebook collection |
Eye candy here generated with tigramite. Just out of scope is the Python Causality Handbook (github)
| Downloads | Hello ? | Remarks, examples | ||
| causalnex | 15,709 | QuantumBlack | ||
| syntheticcontrolmethods | 2,757 | explanation | ||
| lingam | 1,592 | site paper | ||
| tigramite | 412 | notebook gallery | paper | |
| misso | 283 | |||
| nonlincausality | 109 | broken? | ||
| pycute | 88 | |||
| Not on PyPI... | ||||
| rgcpd | ||||
| tcdf | ||||
| AmortizedCausalDiscovery | paper | |||
| VLTimeSeriesCausality | ||||
| sru_for_cgi | ||||
| exact_linear_dependence | ||||
Eye candy here.
| Downloads | Hello ? | Discuss | |
| copulas | 184,467 | gallery | |
| copulabayesnet | 553 | ||
| cvinemodel | 371 | ||
| pycopula | 296 | ||
| copulalib | 272 | broken? | |
| copula | 110 | ||
| copent | 98 |
If you have a suggestion for these lists, please file an issue.
| Downloads | Hello? | Remarks, examples | Discuss | |
| ruptures | 559,553 | paper | ||
| changefinder | 15,899 | |||
| changepy | 11,419 | |||
| bayesian-changepoint-detection | 3,408 | |||
| changepoint_cython | 475 | from R changepoint | ||
| bocd | 138 | |||
| roufcp | 138 | |||
| changepoint | 109 | paper |
If you have a suggestion for these lists, please file an issue.
| Downloads | Hello ? | Remarks, examples | Discuss | |
| tsfresh | 168,000 | features | ||
| ta | 91,000 | technical indicators | ||
| pandas-ta | 59,089 | technical indicators | ||
| liminol | 40,725 | kernel | ||
| traces | 39,432 | irregular | ||
| stationarizer | 647 | |||
| mvtsdata_toolkit | 600 | bitbucket | ||
| luminaire | 500 | |||
| magi | 480 | dask | ||
| changepoint_cython | 475 | from R changepoint | ||
| bocd | 138 | change point | ||
| roufcp | 138 | change point | ||
| changepoint | 109 | paper | ||
| pytsmp | 47 | |||
| The following are not on PyPI | ||||
| anomalyMulti | blog | |||
| db_oc_tsc | paper | fe_ge_tsc (features) | ||
| khiva-python |
See also previous category and list of papers. If you have a suggestion for these lists, please file an issue.
| Downloads | Hello ? | Remarks, examples | Discuss | |
| pywavelets | 2,219,206 | |||
| tslearn | 50,000 | |||
| stumpy | 21,000 | notebook | post | |
| signatory | 9,952 | paper | ||
| pyEDM | 9,328 | |||
| sktime | 8,864 | paper classification | ||
| matrixprofile | 5,839 | |||
| matrixprofile-ts | 3,002 | |||
| pyts | 2,584 | |||
| saxpy | 1,479 | paper | ||
| mcfly | 658 | |||
| kymatio | 482 | wavelets | ||
| knn-lbenhanced | 56 | paper | ||
| The following are not on PyPI | ||||
| rocket | paper | |||
| mkalgo | notebook | (fork) | ||
| os-cnn | paper |
New category. Needs improvement.
| Downloads | Hello? | Remarks, examples | Discuss | |
| deepecho | 3,200 | notebook | new! | discussion |
| timesynth | 166 | |||
| tsBNgen | 98 | notebook | paper | |
| data-generation | 76 | |||
| The following are not released on PyPI | ||||
| tsgan | ML-AIM Lab | |||
| synthetic-data-for-finance | ||||
| timegan | paper | |||
| TimeSeries-GAN | ||||
| random-timeseries-generator |
If you have a suggestion for these lists, please file an issue.
New category
| Downloads | Hello ? | Remarks, examples | |
| microprediction | 7000 | ||
| timexseries | 325 | FAAS | |
| automlservice | TPOT, flask |
New category
| Downloads | Hello ? | Remarks, examples | |
| pycausalimpact | 11,032 | ||
| causallib | 1,689 |
See also scipy, statsmodels etc
| Downloads | Notes | |
| tdigest | 432,000 | Online. See blog article. |
| iminuit | 25,122 | Robust likelihood distributions |
| fitter | 9,806 | |
| powerlaw | 6,128 | |
| reliability | 5,906 | Censored |
There's a bit of a fine line between distribution fitting packages and optimization.
I've moved this table into HumpDay: A Package to Help You Choose a Python Global Optimizer. As you can read there, the following are current recommendations - probably better maintained on that page than here.
| Package | Examples | Form Guide |
| dlib | dlib | The algorithm find_min_global is fast, light and ingenious. It estimates the Lipschitz constant and constructs linear upper bounds, then samples the upper envelope function. If the degree to which your objective function varies across space isn't too great, and you are in dim<25, this is one way to get a free lunch. You will kick yourself for not inventing this. Strong contender! |
| pySOT | dycors; srbf | Implements Regis and Shoemaker's method using radial basis surrogates (paper), and stochastic radial basis function (paper). Compared to dlib, it is very heavy on computation, but if you are drilling holes in the ground that isn't going to be an issue. Definitely include in your trifectas. |
| skopt | default | skopt.gp_minimize. Like pySOT this uses surrogates and is good but by necessity, computationally heavy. A very popular library and underrepresented in the current Elo ratings somewhat due to the paucity of my wrapper. Expect good things. |
| nevergrad | ngpopt8 ; ngpopt4 | A product of Facebook research, ngopt8 selects between optimizers, guided by benchmarking that is probably better than mine. See Algorithm 1 page 3 of the paperBlack Box Optimization Revisited by Meunier et al. I'm reliably informed that new meta-optimizers are on the way - yet the latest model is topping some of the leaderboards already. Don't lay. |
| shgo | nelder | As discussed in Comparing Global Optimizers, SHGO is now one of the scipy global searches. It uses techniques from simplicial homology to avoid wasting samples, and seems to work very well in relatively low dimensional problems (say <=10) using either Nelder-Mead or Powell for local search. Each way chance. |
| pymoo | pattern | Hooke and Jeeves pattern search (package notes, paper) from 1960 is still pretty good! Don't overlook. |
| ax-platform | default | Another effort at Facebook that is independent of nevergrad. The default global optimizer is part of a more extensive suite of tools and I'm reliably informed that some integration with nevergrad is likely (notwithstanding this issue) Expect big things. |
| optuna | cmaes | I often see optuna compared to hyperopt, as if those were the only two packages in town. Well, if you must choose one of those two, make it Optuna. Flexible and well supported. A good bet in the right field. |
Some were compared in Comparing Python Global Optimizers previously. You can very marginally help this effort by yelling and screaming on Linked-in about it or joining discussions listed here, should you feel so inclined. Better yet, put a PR into HumpDay. There's also a library discussed here that you might want to try in conjunction with these. It modifies the objective function, so is uncoupled from the optimizer itself.
Equities and crypto mostly
| Downloads | Notes | ||
| backtrader | 15,023 | crypto | |
| prophet | 10,872 | ||
| octobot | 9,542 | ||
| jesse | 5,012 | crypto | |
| bt | 4,563 | fixed income | |
| vectorbt | 4,192 | ||
| backtesting | 4,182 | ||
| finlab-crypto | 4,179 | ||
| zipline | 4,178 | ||
| qtpylib | 4,064 | ||
| rqalpha | 2,260 | chinese doc | |
| pyalgotrade | 900 | ||
| zvt | 900 | chinese doc | |
| btrccts | 880 | crypto | |
| qstrader | 426 | ||
| quant | 280 | ||
| pinkfish | 157 | ||
| quantdom | 174 |
If you have a suggestion for these lists, please file an issue.
New category
| The following are not deployed on PyPI | ||
| pategan | paper |
Some potentially underestimated statistical libraries.
| Downloads | Remarks | |||
| pycaret | 220,000 | now with ts | ||
| combo | 86,774 | emsembles | ||
| suod | 86,215 | outliers | ||
| pyod | 85,000 | outliers | ||
| pycaret | 53,000 | time-series article | ||
| hyppo | 3,400 | multivariate hypothesis | ||
| baycomp | 3,300 | classification evaluation | ||
| model_search | Feb 19, 2021 |
To ensure articles like this are in your thread, or category updates here consider following microprediction on LinkedIn. We're trying to make powerful bespoke AI free to the world, and you are welcome to contribute in large or small ways. If you have a suggestion for these lists, please file an issue.
Comments