Lightweight univariate time-series forecasting with auto-tuned models — NumPy only.

These details have not been verified by PyPI

Project description

randomstatsmodels

Check out medium story here: Medium Story

Lightweight univariate time-series forecasting with auto-tuned models — NumPy only, no heavy dependencies.

19 forecasting models with a unified .fit(y) / .predict(h) API. Each Auto* wrapper grid-searches hyperparameters on a validation split and refits the best configuration on the full series.

Installation

pip install randomstatsmodels

Requires: Python 3.9+ and NumPy.

Quick Start

from randomstatsmodels import AutoKoopman, AutoPolymath, AutoSSA
import numpy as np

rng = np.random.default_rng(42)
t = np.arange(200)
y = 10 + 0.05 * t + np.sin(2 * np.pi * t / 24) + 0.1 * rng.normal(size=t.size)

h = 12  # forecast horizon

model = AutoKoopman().fit(y)
yhat = model.predict(h)
print("Forecast:", yhat[:5])
print("Best config:", model.best_)

Models

19 models organised by approach. Every Auto* class accepts a parameter grid, fits/evaluates candidates, and exposes .fit(y) and .predict(h).

Classical

Model	Description
`AutoNaive`	Baselines: last value, seasonal, drift, mean
`AutoHoltWinters`	Triple exponential smoothing (level + trend + seasonal)
`AutoThetaAR`	Theta method with AR(1) residual correction
`AutoFourier`	Harmonic regression with optional linear trend

Regression-Based

Model	Description
`AutoNEO`	Nonlinear Evolution Operator — polynomial AR features
`AutoPolymath`	Polynomial + Fourier basis + ridge regression
`AutoLocalLinear`	Weighted local regression with exponential decay
`AutoPALF`	Proximal Aggregation Lag Forecaster — penalised lag weighting

Decomposition / Spectral

Model	Description
`AutoSSA`	Singular Spectrum Analysis — SVD on trajectory matrix
`AutoKoopman`	Dynamic Mode Decomposition / Koopman operator via delay embedding
`AutoSpectralGradient`	Spectral derivative flow — extrapolates Fourier mode dynamics

Advanced / Calculus-Based

Model	Description
`AutoFracDiff`	Fractional calculus — Grunwald-Letnikov fractional differencing + AR
`AutoGreensKernel`	Integral equation — Green's function convolution kernel
`AutoPDEField`	Partial differential equations — advection-diffusion on time-scale field
`AutoVariationalPath`	Calculus of variations — Euler-Lagrange optimal path

Hybrid / Meta

Model	Description
`AutoHybridForecaster`	Linear (Fourier + trend + AR) + GRU residual network
`AutoMELD`	Multiscale embedding with Random Fourier Features + kNN
`AutoRIFT`	Recursive Information Flow Tensor — information-channel dynamics
`AutoEnsemble`	Combines multiple base forecasters with learned weights

Advanced Ensembles

Model	Description
`AutoStacked`	Meta-learner stacking — ridge regression on base model predictions
`AutoBagged`	Block-bootstrap bagging — median of models trained on resampled series
`AutoDynamic`	Horizon-adaptive weighting — model weights change per forecast step

Benchmarks

22 models evaluated on 20 real-world time series (10 from FRED with 300-600 points, 10 classic hardcoded) using speed="slow" grids. Ranked by RMSE per dataset. Includes 5 statsforecast baselines. AutoRIFT and AutoHybridForecaster excluded from fast benchmark (>30s fit time per dataset).

Overall Rankings

Rank	Model	Type	Avg Rank	#1st	#Top3	#Top7	Median MAE	Median RMSE	Median MAPE	Median sMAPE
1	SF_AutoTBATS	statsforecast	7.30	1	5	10	18.28	23.53	7.66%	7.77%
2	SF_AutoARIMA	statsforecast	8.00	1	3	10	18.03	23.29	7.14%	7.65%
3	AutoNEO	randomstatsmodels	8.94	2	4	8	25.86	36.51	7.38%	7.59%
4	AutoMELD	randomstatsmodels	8.95	2	4	7	20.80	27.83	7.79%	7.84%
5	SF_AutoETS	statsforecast	9.35	0	0	9	33.65	38.18	11.53%	12.08%
6	SF_AutoCES	statsforecast	9.80	5	6	9	18.96	23.27	10.70%	10.71%
7	AutoNaive	randomstatsmodels	10.05	1	1	7	24.43	30.87	12.94%	13.68%
8	AutoKoopman	randomstatsmodels	10.75	0	5	5	31.60	40.47	12.63%	12.59%
9	AutoPolymath	randomstatsmodels	10.84	0	4	6	22.65	29.33	9.48%	9.00%
10	SF_AutoTheta	statsforecast	11.00	0	3	7	20.07	25.67	8.97%	9.27%
11	AutoPALF	randomstatsmodels	11.00	2	2	7	42.30	49.53	12.77%	13.84%
12	AutoKNN	randomstatsmodels	11.05	1	2	8	35.92	43.24	12.71%	13.89%
13	AutoPDEField	randomstatsmodels	11.25	0	2	5	36.12	41.58	11.16%	11.66%
14	AutoSSA	randomstatsmodels	11.75	0	3	6	31.47	38.00	12.24%	13.28%
15	AutoThetaAR	randomstatsmodels	11.80	0	1	5	38.03	43.81	15.11%	16.76%
16	AutoHoltWinters	randomstatsmodels	12.20	2	2	7	49.16	60.66	14.24%	14.63%
17	AutoGreensKernel	randomstatsmodels	12.60	1	3	6	40.53	48.61	15.87%	16.51%
18	AutoVariationalPath	randomstatsmodels	12.80	0	3	5	57.52	67.07	13.10%	13.52%
19	AutoSpectralGradient	randomstatsmodels	13.25	1	3	4	54.54	64.10	15.53%	15.02%
20	AutoFourier	randomstatsmodels	14.15	1	3	4	69.65	83.66	22.71%	22.45%
21	AutoLocalLinear	randomstatsmodels	15.10	0	1	5	56.44	63.38	20.47%	25.01%
22	AutoFracDiff	randomstatsmodels	18.60	0	0	0	110.80	113.95	30.43%	36.00%

Speed Presets

Every Auto* model accepts a speed parameter controlling grid search thoroughness:

from randomstatsmodels import AutoKoopman

model = AutoKoopman(speed="super_fast")   # ~1 combo, seconds
model = AutoKoopman(speed="fast")         # ~6 combos
model = AutoKoopman(speed="normal")       # default grids
model = AutoKoopman(speed="slow")         # ~864 combos
model = AutoKoopman(speed="super_slow")   # ~2640 combos

Datasets

20 real-world datasets (10 from FRED with 300-600 points, 10 classic):

Source	Datasets
FRED (300-600 pts)	USUnemployment, USConsumerPrices, USIndProdIndex, USHousingStarts, USMoneySupply, FedFundsRate, USRetailSales, USElectricity, SP500, US10YrTreasury
Classic (89-168 pts)	MilkProduction, GoldPrice, NileMinLevel, AirPassengers, CO2, GlobalTemp, IntlAirline, Lynx, Nile, Sunspots

Key Findings

SF_AutoTBATS is #1 by RMSE — strong on seasonal FRED data
SF_AutoARIMA is #2 — most consistent across dataset types
AutoNEO is the best randomstatsmodels model (#3) — polynomial AR features
AutoMELD is #4 — multiscale embedding excels on longer series
AutoKoopman (#8) — DMD eigenvalue propagation, 5 top-3 finishes
No single model dominates — model selection matters for your data type

Per-Model Speed Benchmarks (100 real-world datasets)

AutoNaive

Speed	Mean MAE	Mean RMSE	Mean sMAPE	Median MAE	Median RMSE	Median sMAPE	OK
super_fast	48402.79	83881.03	32.56%	41.87	51.48	20.43%	100
fast	46825.06	82219.58	31.01%	39.12	47.55	19.36%	100
normal	55665.87	82470.84	29.79%	27.33	33.06	17.35%	100
slow	55665.62	82470.65	29.85%	33.33	40.96	16.32%	100
super_slow	55666.08	82471.08	29.84%	33.33	40.96	16.32%	100

AutoThetaAR

Speed	Mean MAE	Mean RMSE	Mean sMAPE	Median MAE	Median RMSE	Median sMAPE	Median Fit (s)	OK
super_fast	66231.58	105091.88	28.39%	38.03	43.80	15.98%	0.01	100
fast	66231.58	105091.88	28.39%	38.03	43.80	15.98%	0.01	100
normal	66231.58	105091.88	28.39%	38.03	43.80	15.98%	0.01	100
slow	66231.58	105091.88	28.39%	38.03	43.80	15.98%	0.01	100
super_slow	66231.58	105091.88	28.39%	38.03	43.80	15.98%	0.01	100

AutoKNN

Speed	Mean MAE	Mean RMSE	Mean sMAPE	Median MAE	Median RMSE	Median sMAPE	Median Fit (s)	OK
super_fast	45495.17	81223.33	30.94%	37.34	43.61	17.77%	0.01	100
fast	46081.70	81754.65	32.17%	39.10	45.58	19.54%	0.04	100
normal	44846.01	80618.64	32.61%	38.31	47.24	19.55%	0.06	100
slow	45077.83	81224.90	31.70%	40.49	51.34	18.16%	0.19	99
super_slow	45711.61	81671.07	30.70%	40.49	51.34	18.02%	0.60	99

AutoPDEField

Speed	Mean MAE	Mean RMSE	Mean sMAPE	Median MAE	Median RMSE	Median sMAPE	Median Fit (s)	OK
super_fast	71364.72	113802.04	34.27%	46.73	54.63	16.75%	0.00	100
fast	62836.93	100782.31	34.26%	46.73	54.63	16.66%	0.01	100
normal	61394.96	98643.02	32.35%	30.15	36.51	15.19%	0.14	100
slow	59601.86	96607.74	32.43%	30.10	36.19	14.63%	0.22	100
super_slow	59880.46	96029.28	31.13%	30.90	36.42	14.23%	3.22	100

AutoFourier

Speed	Mean MAE	Mean RMSE	Mean sMAPE	Median MAE	Median RMSE	Median sMAPE	Median Fit (s)	OK
super_fast	55563.07	80172.78	33.44%	57.76	68.20	22.97%	0.00	100
fast	51677.30	76308.97	36.55%	57.76	68.20	23.72%	0.00	100
normal	52139.33	79995.06	35.59%	61.08	66.96	23.36%	0.00	100
slow	49132.03	77423.79	35.22%	61.08	66.96	23.36%	0.01	100
super_slow	48872.80	77122.95	33.60%	61.08	66.96	23.36%	0.02	100

AutoGreensKernel

Speed	Mean MAE	Mean RMSE	Mean sMAPE	Median MAE	Median RMSE	Median sMAPE	Median Fit (s)	OK
super_fast	72348.83	99254.58	51.74%	111.93	126.21	35.64%	0.01	100
fast	70411.59	96782.30	50.11%	104.02	112.41	34.89%	0.03	100
normal	50902.81	77641.40	34.90%	47.15	56.76	22.73%	0.21	100
slow	49221.30	76075.91	34.15%	45.95	55.78	21.83%	0.61	100
super_slow	49778.84	78193.66	33.07%	44.38	54.47	20.69%	6.14	100

AutoSpectralGradient

Speed	Mean MAE	Mean RMSE	Mean sMAPE	Median MAE	Median RMSE	Median sMAPE	Median Fit (s)	OK
super_fast	52722.21	84979.30	33.96%	58.04	67.81	19.47%	0.00	100
fast	44047.84	74975.53	33.86%	51.63	63.20	19.38%	0.01	100
normal	51565.84	80725.74	33.57%	49.33	60.86	18.60%	0.32	100
slow	59388.87	86750.79	35.76%	53.36	64.10	21.49%	0.92	100
super_slow	59750.35	86300.03	36.95%	52.61	58.78	21.71%	13.32	100

AutoFracDiff

Speed	Mean MAE	Mean RMSE	Mean sMAPE	Median MAE	Median RMSE	Median sMAPE	Median Fit (s)	OK
super_fast	135625.12	165178.89	85.97%	501.87	571.23	86.19%	0.01	100
fast	130154.81	157580.15	81.32%	455.90	483.54	81.51%	0.02	100
normal	95424.54	126662.35	56.29%	254.84	256.02	51.90%	0.30	100
slow	85441.64	119020.80	38.24%	111.51	115.23	31.00%	1.13	100
super_slow	87305.06	121693.81	36.59%	94.22	98.15	27.19%	7.01	100

AutoKoopman

Speed	Mean MAE	Mean RMSE	Mean sMAPE	Median MAE	Median RMSE	Median sMAPE	Median Fit (s)	OK
super_fast	45411.65	72908.35	33.11%	35.13	43.04	19.67%	0.00	100
fast	44888.15	72263.73	33.14%	34.39	41.02	21.42%	0.01	100
normal	52062.77	81243.92	29.71%	29.11	34.77	18.06%	0.77	100
slow	52822.65	82219.21	30.71%	28.06	34.25	18.18%	3.75	100
super_slow	50844.52	79799.29	28.94%	35.61	44.43	17.06%	28.62	100

AutoLocalLinear

Speed	Mean MAE	Mean RMSE	Mean sMAPE	Median MAE	Median RMSE	Median sMAPE	Median Fit (s)	OK
super_fast	80495.97	113829.66	35.81%	33.08	40.21	15.17%	0.01	100
fast	45131.52	70541.57	32.41%	28.64	35.57	15.60%	0.01	100
normal	75742.64	104015.39	39.67%	31.57	37.60	15.31%	0.03	100
slow	124413.16	151654.08	48.26%	31.57	37.60	18.08%	0.06	100
super_slow	168910.25	206433.85	49.40%	46.44	52.03	25.45%	0.12	100

AutoNEO

Speed	Mean MAE	Mean RMSE	Mean sMAPE	Median MAE	Median RMSE	Median sMAPE	Median Fit (s)	OK
super_fast	52987.72	87972.01	26.21%	32.66	37.92	10.66%	0.00	100
fast	56911.65	91148.54	26.16%	27.26	36.40	10.95%	0.03	96
normal	63926.05	101068.48	27.53%	30.52	38.01	13.53%	1.44	88
slow	68197.11	112259.65	29.88%	24.71	30.17	12.98%	9.79	86
super_slow	67755.89	109997.50	32.60%	19.14	26.63	17.18%	32.11	89

AutoPolymath

Speed	Mean MAE	Mean RMSE	Mean sMAPE	Median MAE	Median RMSE	Median sMAPE	Median Fit (s)	OK
super_fast	49431.52	81177.78	26.29%	33.47	38.35	11.78%	0.00	100
fast	67358.56	102976.44	27.13%	27.62	32.61	11.43%	0.05	95
normal	83892.45	139210.80	27.52%	24.51	32.43	12.03%	2.97	88
slow	84919.11	137332.12	29.43%	24.29	32.35	12.59%	6.15	90
super_slow	188379.08	377591.15	29.79%	34.81	41.63	13.12%	31.60	84

AutoSSA

Speed	Mean MAE	Mean RMSE	Mean sMAPE	Median MAE	Median RMSE	Median sMAPE	Median Fit (s)	OK
super_fast	116289.21	160439.84	32.37%	43.25	46.83	18.16%	0.39	100
fast	129346.18	172263.47	35.45%	33.07	43.96	19.75%	0.67	100
normal	149202.82	198303.70	36.96%	34.21	45.90	18.28%	2.21	100
slow	136224.02	189953.53	36.68%	37.02	46.50	19.44%	4.28	100
super_slow	107759.60	152480.05	35.44%	34.69	41.52	20.48%	11.28	100

AutoPALF

Speed	Mean MAE	Mean RMSE	Mean sMAPE	Median MAE	Median RMSE	Median sMAPE	Median Fit (s)	OK
super_fast	47702.41	83258.51	31.67%	45.66	53.44	19.58%	0.68	100
fast	47683.64	83242.77	32.32%	44.63	52.34	19.58%	1.37	100
normal	46707.61	82428.52	31.62%	43.89	52.19	19.58%	9.08	100
slow	47864.12	83421.89	31.33%	43.88	52.16	19.57%	11.87	100
super_slow	47862.15	83421.78	31.18%	43.87	52.16	19.57%	25.71	100

AutoVariationalPath

Speed	Mean MAE	Mean RMSE	Mean sMAPE	Median MAE	Median RMSE	Median sMAPE	Median Fit (s)	OK
super_fast	78852.69	102207.70	55.65%	92.87	104.41	43.88%	0.01	100
fast	63295.11	86612.91	33.31%	54.67	65.03	20.51%	0.03	100
normal	66359.42	91448.78	33.37%	38.91	46.99	18.43%	1.64	100
slow	61857.70	87102.42	32.95%	39.46	45.33	18.98%	6.65	100
super_slow	57655.57	83079.29	32.50%	39.36	45.26	18.04%	42.95	100

Metrics

from randomstatsmodels.metrics import mae, mse, rmse, mape, smape

The evaluation framework also provides MASE, MSSE, and Median Absolute Error.

License

MIT

Project details

These details have not been verified by PyPI

Release history Release notifications | RSS feed

This version

4.5.1

Mar 27, 2026

4.5.0

Mar 26, 2026

4.1.0

Mar 26, 2026

4.0.7

Mar 26, 2026

4.0.6

Mar 26, 2026

4.0.5

Mar 26, 2026

4.0.4

Mar 26, 2026

4.0.3

Mar 26, 2026

4.0.2

Mar 26, 2026

4.0.1

Mar 26, 2026

4.0.0

Mar 26, 2026

3.3.1

Mar 25, 2026

3.3.0

Mar 25, 2026

3.2.1

Mar 25, 2026

3.2.0

Mar 25, 2026

3.1.1

Mar 25, 2026

3.1.0

Mar 25, 2026

3.0.0

Mar 24, 2026

2.0.0

Mar 24, 2026

1.6.1

Feb 1, 2026

1.6.0

Feb 1, 2026

1.5.0

Feb 1, 2026

1.4.0

Feb 1, 2026

1.3.0

Feb 1, 2026

1.2.0

Feb 1, 2026

1.1.2

Aug 29, 2025

1.1.1

Aug 29, 2025

1.1.0

Aug 29, 2025

1.0.3

Aug 28, 2025

1.0.2

Aug 28, 2025

1.0.1

Aug 28, 2025

1.0.0

Aug 28, 2025

0.1.0

Aug 27, 2025

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

randomstatsmodels-4.5.1.tar.gz (112.4 kB view details)

Uploaded Mar 27, 2026 Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

The dropdown lists show the available interpreters, ABIs, and platforms. Enable javascript to be able to filter the list of wheel files.

randomstatsmodels-4.5.1-py3-none-any.whl (122.7 kB view details)

Uploaded Mar 27, 2026 Python 3

File details

Details for the file randomstatsmodels-4.5.1.tar.gz.

File metadata

Download URL: randomstatsmodels-4.5.1.tar.gz
Upload date: Mar 27, 2026
Size: 112.4 kB
Tags: Source
Uploaded using Trusted Publishing? No
Uploaded via: twine/6.2.0 CPython/3.11.1

File hashes

Hashes for randomstatsmodels-4.5.1.tar.gz
Algorithm	Hash digest
SHA256	`16159bafacc023b2ff06d089e5d91d3b1cf740824302220ff91022a82c2a0765`
MD5	`dcdd29c214056c1098bbf2fc76dcb318`
BLAKE2b-256	`d475bf79555fa238ff565ff6c9282c4c94bbe58dd45de85ec7a5d5491740f43b`

See more details on using hashes here.

File details

Details for the file randomstatsmodels-4.5.1-py3-none-any.whl.

File metadata

Download URL: randomstatsmodels-4.5.1-py3-none-any.whl
Upload date: Mar 27, 2026
Size: 122.7 kB
Tags: Python 3
Uploaded using Trusted Publishing? No
Uploaded via: twine/6.2.0 CPython/3.11.1

File hashes

Hashes for randomstatsmodels-4.5.1-py3-none-any.whl
Algorithm	Hash digest
SHA256	`5eae0468b03e856e71c3c33bf5bb88da271be8c7fd8bced0a5c7ef0197a224c8`
MD5	`e59da67a142fd24e6529186db144a7d8`
BLAKE2b-256	`3136fbe6b80f5a554bd6b7984abcd2a7c20755e1b763073ff4c6a21c6d7a370b`

See more details on using hashes here.

randomstatsmodels 4.5.1

Navigation

Verified details

Maintainers

Unverified details

Meta

Classifiers

Project description

randomstatsmodels

Installation

Quick Start

Models

Classical

Regression-Based

Decomposition / Spectral

Advanced / Calculus-Based

Hybrid / Meta

Advanced Ensembles

Benchmarks

Overall Rankings

Speed Presets

Datasets

Key Findings

Per-Model Speed Benchmarks (100 real-world datasets)

AutoNaive

AutoThetaAR

AutoKNN

AutoPDEField

AutoFourier

AutoGreensKernel

AutoSpectralGradient

AutoFracDiff

AutoKoopman

AutoLocalLinear

AutoNEO

AutoPolymath

AutoSSA

AutoPALF

AutoVariationalPath

Metrics

License

Project details

Verified details

Maintainers

Unverified details

Meta

Classifiers

Release history Release notifications | RSS feed

Download files

Source Distribution

Built Distribution

File details

File metadata

File hashes

File details

File metadata

File hashes