{% extends "main.html" %} {% block tabs %} {{ super() }} {% endblock %} {% block content %}
v0.0.3 — Rust Turbo Mode Available

Time Series Forecasting
Made Simple

One line of code. 30+ models. Zero configuration.
Pure Python — no compiled dependencies required.

30+ Models
3 Dependencies
10x Rust Turbo
387 Tests
Install Quickstart GitHub
forecast.py
from vectrix import forecast

result = forecast("sales.csv", steps=12)
result.plot() # auto model + intervals + plot
200K+ Time series benchmarked on M3 & M4 competitions
65+ Features extracted by Forecast DNA fingerprinting
0.696 M4 Hourly OWA — competition winner level accuracy
<5ms ETS model fit with Rust turbo acceleration
Features

Everything you need for forecasting

From one-line forecasts to advanced adaptive intelligence — Vectrix scales with your needs.

Zero Configuration

Pass your data, get forecasts. Auto model selection, hyperparameter tuning, and flat-line defense — all automatic.

30+ Statistical Models

ETS, ARIMA, Theta, MSTL, TBATS, GARCH, CES, Croston, DOT, 4Theta, ESN, DTSF — and more.

Flat Defense System

Unique 4-level system prevents constant predictions — the most common forecasting failure mode in production.

Forecast DNA

65+ feature fingerprinting. Know your data's difficulty, optimal model, seasonality, and similarity to other series.

Adaptive Intelligence

Regime detection (HMM), self-healing forecasts (CUSUM + EWMA), and 8 business constraint types.

Pure Python

Built on NumPy + SciPy + Pandas only. No compiled extensions, no platform issues. Optional Rust turbo for 10x speed.

Probabilistic Forecasting

Parametric distributions (Gaussian, Student-t, Log-Normal), quantile forecasts, conformal prediction intervals, CRPS scoring.

R-Style Regression

Formula interface: regress(df, "sales ~ ads + price"). OLS, Ridge, Lasso, Huber, Quantile with full diagnostics.

Business Intelligence

Anomaly detection, what-if scenarios, backtesting framework, and production metrics (WAPE, MASE, bias).

How it works

Three lines. Full pipeline.

Forecasting, analysis, and regression — each designed to work with one function call.

Forecasting
from vectrix import forecast

result = forecast("sales.csv", steps=12)
result.plot()
result.to_csv("forecast.csv")
Analysis
from vectrix import analyze

report = analyze("data.csv")
print(report.dna.difficulty)
print(report.dna.recommended_model)
Regression
from vectrix import regress

r = regress(df, "sales ~ ads + price")
print(r.r_squared)
r.plot()
Model Catalog

30+ models, one unified API

Every model shares the same fit → predict interface. Vectrix picks the best one for your data automatically.

Exponential Smoothing
AutoETS Holt-Winters SES DES AutoCES 4Theta
ARIMA Family
AutoARIMA ARIMAX SARIMA
Decomposition
AutoMSTL AutoTBATS Theta DOT
Non-Parametric
DTSF ESN Fourier
Volatility & Special
GARCH AutoCroston ADIDA IMAPA
ML & Foundation
LightGBM XGBoost Chronos-2 TimesFM
Adaptive Intelligence

Forecasts that adapt to reality

Real-world data is messy. Regimes shift, anomalies appear, business rules matter. Vectrix handles it all.

Regime Detection

HMM-based automatic regime identification. Detect structural breaks and adapt model selection accordingly.

Self-Healing Forecasts

CUSUM + EWMA monitoring. When predictions drift, the system automatically corrects before errors accumulate.

8 Business Constraints

Non-negative, bounded range, monotonic, integer-only, seasonal floor/ceiling, growth rate limits, and more.

Forecast DNA Fingerprinting

65+ features: trend strength, seasonality type, noise ratio, Hurst exponent, demand density, and more. Know your data before you model it.

adaptive.py
from vectrix import Vectrix

vx = Vectrix(data)

# regime detection
regimes = vx.detectRegimes()

# self-healing forecast
healed = vx.healForecast(
  steps=12,
  actual=newData
)

# business constraints
result = vx.forecast(
  steps=12,
  constraints=[
    NonNegative(),
    MaxGrowth(0.1)
  ]
)
Performance

Rust Turbo Mode

Optional Rust-accelerated core loops. Pre-built wheels for all platforms — no Rust compiler needed.

AutoETS
Python: 348ms
Rust: 32ms
10.8x faster
AutoARIMA
Python: 285ms
Rust: 51ms
5.6x faster
Theta
Python: 130ms
Rust: 16ms
8.1x faster

pip install "vectrix[turbo]" — automatic fallback to pure Python if wheels are unavailable.

Benchmarks

M3 & M4 Competition Results

Tested against 200,000+ real-world time series. OWA < 1.0 means beating the Naive2 baseline.

Competition Yearly Quarterly Monthly Weekly Daily Hourly
M3 0.848 0.825 0.758 — — 0.819
M4 0.974 0.797 0.987 0.737 1.207 1.006
M4 Ensemble 0.879 0.797 0.927 0.737 1.105 0.696

M4 Ensemble uses VX-Ensemble with DOT + AutoCES + 4Theta + DTSF + ESN. Hourly 0.696 OWA = competition winner level.

View full benchmark results →

pipeline.py
from vectrix import Vectrix

vx = Vectrix(data)

# full pipeline
pipe = vx.pipeline([
  "log",
  "deseasonalize",
  "scale"
])
result = pipe.forecast(12)

# backtesting
bt = vx.backtest(
  steps=12,
  windows=5
)

# what-if scenarios
scenarios = vx.whatIf(
  shocks=[-0.2, 0, 0.2]
)
Advanced Features

Beyond basic forecasting

When you need more control, Vectrix gives you the full toolkit.

sklearn-Style Pipeline

Chain transformers (log, scale, deseasonalize) with automatic inverse on predictions. Fully serializable.

Backtesting Framework

Sliding window cross-validation with customizable metrics. Know how your model performs before deployment.

What-If Scenarios

Simulate shock impacts on forecasts. Perfect for business planning and stress testing.

Multivariate (VAR/VECM)

Vector autoregression with Granger causality, impulse response, and cointegration testing.

Use Cases

Built for real-world applications

From demand planning to financial modeling — Vectrix adapts to your domain.

Retail & E-Commerce

Demand forecasting, inventory optimization, seasonal planning with intermittent demand support (Croston, ADIDA).

Finance & Trading

Volatility modeling (GARCH), regime detection for market states, probabilistic risk assessment with confidence intervals.

Energy & Utilities

Load forecasting with multi-seasonal patterns (hourly/daily/weekly). MSTL and TBATS handle complex seasonality.

Healthcare & IoT

Patient volume prediction, sensor data forecasting, anomaly detection with self-healing for drift correction.

HR & Workforce

Staffing demand prediction, attrition forecasting. Hierarchical reconciliation for department-level planning.

SaaS & Product

User growth modeling, revenue forecasting, churn prediction. Business constraints enforce realistic growth bounds.

Get Started

Install in seconds

Choose the installation that fits your needs.

$ pip install vectrix
Core — NumPy + SciPy + Pandas. All 30+ models included.
Turbo — Add Rust acceleration. 5-10x faster. Pre-built wheels.
ML — Add LightGBM + XGBoost + scikit-learn models.
Foundation — Add Chronos-2 + TimesFM zero-shot models.
Comparison

How Vectrix compares

Purpose-built for production forecasting with features no other library offers.

Capability Vectrix statsforecast Prophet Darts
Zero-config forecasting Yes Yes No No
Pure Python (no compiled deps) Yes No No No
30+ statistical models Yes Yes No Yes
Flat prediction defense Yes No No No
Forecast DNA fingerprinting Yes No No No
Business constraints (8 types) Yes No No No
R-style regression Yes No No No
Self-healing forecasts Yes No No No
Foundation model wrappers Yes No No Yes
VAR / VECM multivariate Yes No No Yes
Open Source

MIT Licensed. Free forever.

Use Vectrix freely in personal and commercial projects. Contributions welcome.

MIT License

No restrictions on commercial use

GitHub

Star, fork, and contribute

387 Tests

Comprehensive test coverage

Full Docs

API reference, guides, and tutorials

Ready to forecast?

Get started with Vectrix in under a minute.

Get Started Star on GitHub Buy Me A Coffee Buy me a coffee
{% endblock %}