pypricing

Bayesian log-demand pricing analytics with PyMC for long-format panels (one row per SKU–time observation): own-price elasticity, optional controls, instruments, hierarchy, trend/seasonality, and counterfactual prediction.

Install

pip install pypricing

model.graphviz() also needs the system Graphviz binaries (e.g. brew install graphviz on macOS).

Full docs (guides + API + notebooks): pypricing.readthedocs.io

Quickstart

import numpy as np

from pypricing import LogLogDemandModel, generate_mock_data

df = generate_mock_data(
    n_periods=20,
    n_skus=5,
    n_controls=2,
    include_seasonality=False,
    random_state=0,
)

model = LogLogDemandModel()
model.fit(df, draws=1000, tune=1000, chains=4, random_seed=42)

df_scenario = df.drop(columns=["quantity"]).copy()
df_scenario["price"] = df_scenario["price"] * 1.05
print(model.predict(df_scenario, hdi_prob=0.9, random_seed=123).head())

Data contract

fit(df) expects level-scale columns:

• Required: sku, price (strictly positive), quantity (non-negative)
• Optional: control_*, iv_*, hierarchy via PanelColumns(group_columns=...), period / region (datetime period required for trend / seasonality)

Column names are configurable via PanelColumns. Internally the model uses log(price) and log(max(quantity, quantity_floor)) (default floor 1.0).

When to use pypricing (and when not to use it)

pypricing assumes products have posted prices that changed over time, and you observe the units sold at each price.

Retail, consumer goods and e-commerce are good use cases: many products, regular sales volume, frequent price or promo changes.

Bad use cases include one-off prices, very low volume, or sales limited by capacity (hotels, airlines).