greybox

Python port of the R greybox package — a toolbox for regression model building and forecasting.

Installation

pip install greybox

For more installation options, see the Installation wiki page.

Quick Example

import numpy as np
import pandas as pd
from greybox import ALM, formula

# Generate sample data
np.random.seed(42)
n = 200
data = pd.DataFrame({
    "y": np.random.normal(10, 2, n),
    "x1": np.random.normal(5, 1, n),
    "x2": np.random.normal(3, 1, n),
})
data["y"] = 2 + 0.5 * data["x1"] - 0.3 * data["x2"] + np.random.normal(0, 1, n)

# Parse formula and fit model
y, X = formula("y ~ x1 + x2", data=data)
model = ALM(distribution="dnorm")
model.fit(X, y)

# Summary
print(model.summary())

# Predict with intervals
pred = model.predict(result.data, interval="prediction", level=0.95)
print(pred.mean[:5])

# Include AR terms (ARIMA-like models)
# For example, ARIMA(1,1,0) model with Log-Normal distribution:
model = ALM(distribution="dlnorm", orders=(1, 1, 0))
model.fit(X, y)

Supported Distributions

Smoothers (lowess, supsmu)

See the Smoothers wiki page for the full reference.

Non-parametric smoothers that reproduce R's stats::lowess and stats::supsmu to machine precision (native pybind11 implementations).

import numpy as np
from greybox import lowess, supsmu

rng = np.random.default_rng(0)
x = np.linspace(0, 6, 80)
y = np.sin(x) + rng.normal(0, 0.2, 80)

# Cleveland's LOWESS — robust local regression
lo = lowess(x, y, f=0.4)
print(lo["x"], lo["y"])   # sorted x, smoothed y

# Friedman's SuperSmoother — variable-span cross-validation
sm_cv    = supsmu(x, y)              # automatic span selection
sm_fixed = supsmu(x, y, span=0.3)    # fixed span

References: Cleveland (1979) for LOWESS, Friedman (1984) for SuperSmoother.

Automatic Identification of Demand (aid, aid_cat)

See the AID wiki page for the full reference.

Classifies a time series into one of six demand types and flags stockouts, new products, and obsolete products. Port of R's greybox::aid() / aidCat().

import numpy as np
import matplotlib.pyplot as plt
from greybox import aid, aid_cat

rng = np.random.default_rng(42)

# Intermittent count demand: Poisson(0.7)
y = rng.poisson(0.7, 120).astype(float)
result = aid(y)
print(result)                       # human-readable summary
print(result.name)                  # e.g. "smooth intermittent count"
print(result.type.type1)            # "count" or "fractional" — R-style attribute access
print(result.type["type1"])         # dict-style fallback also works

# Detect injected stockouts and plot them
y2 = rng.poisson(3, 100).astype(float)
y2[40:50] = 0
result2 = aid(y2)
print(result2.stockouts.start, result2.stockouts.end)  # 1-based, [41] [50]
ax = result2.plot()                 # series + grey-shaded stockout span
plt.show()

# Apply to multiple series at once
series = {
    "a": rng.poisson(1, 80).astype(float),
    "b": rng.poisson(5, 80).astype(float),
    "c": rng.normal(10, 2, 80),
}
cat = aid_cat(series)
print(cat.types)        # 2x3 demand-category frequency table
print(cat.anomalies)    # counts of new / stockouts / old products
ax = cat.plot()         # 2x3 demand-category panel
plt.show()

The nested type and stockouts fields are typed dataclasses that support both result.stockouts.start (R-style attribute access) and result.stockouts["start"] (dict-style fallback).

Features

• ALM (Augmented Linear Model): Likelihood-based regression with 26 distributions
• Formula parser: R-style formulas (y ~ x1 + x2, log(y) ~ ., y ~ 0 + x1) with support for backshift operator
• stepwise(): IC-based variable selection with partial correlations
• CALM(): Combine ALM models based on IC weights
• Forecast error measures: MAE, MSE, RMSE, MAPE, MASE, MPE, sMAPE, and more
• Variable processing: xreg_expander (lags/leads), xreg_multiplier (interactions), temporal_dummy
• Distributions: 27 distribution families with density, CDF, quantile, and random generation
• Association: Partial correlations and measures of association
• Diagnostics: Model diagnostics and validation
• Smoothers: lowess and supsmu matching R's stats::lowess and stats::supsmu to machine precision
• Demand identification: aid() and aid_cat() for automatic classification of demand series and stockout detection
• Method comparison (RMCB): rmcb() for the Nemenyi/MCB test comparing forecasting methods, with "mcb" and "lines" plots

Links

• Wiki documentation
• GitHub repository
• R package on CRAN

License

LGPL-2.1

About

greybox is developed and maintained by OpenForecast, a demand forecasting and inventory management consultancy. The package implements the methods we use in our consulting and teach in our training courses.