cube-alchemy 0.1.4

A lightweight, queryable hypercube engine for multidimensional data analysis

Cube Alchemy

A lightweight, queryable hypercube engine for multidimensional analytics on top of pandas.

Why It Matters

Reduce glue code and speed up your analysis so you can focus on insights.

• Speed: Automatic relationship discovery and traversal.

• Simplicity: Declarative queries achieve slicing and dicing in pure Python with less ad‑hoc joins.

• Concistency: Define your queries and use them everywhere with the same logic and filtering behavior.

• Maintainability: Centralized business logic in reusable components.

• Integration: Power fully interactive analytics apps using frameworks like Streamlit or Panel, or expose it to a web client.

Installation

Requires Python 3.8+.

cd 'your_new_project_path'
python -m venv venv
venv\Scripts\activate
pip install cube-alchemy

Basic usage

Transform your collection of pandas DataFrames into a cohesive analytical model in three simple steps:

• Connect your data - Add your pandas DataFrames to a Hypercube (relationships will be created automatically).

• Define your metrics and queries.

• Query with ease - Extract insights.

flowchart LR
  A["Load DataFrames"] --> B["Build Hypercube"]
  B --> C["Define Metrics"]
  C --> D["Define Queries"]
  D --> E["Execute Queries"]
  E --> F["Update Context State (Apply or Remove Filters)"]
  F --> E

Cube Alchemy connects your data by identifying common column names between DataFrames. These shared columns (edges) form the relationships; automatically building bridges between tables (nodes), handling composite keys when needed. The result is a unified schema you can slice and dice and query in a simple and intuetive way.

import pandas as pd
from cube_alchemy import Hypercube

# 1) Define DataFrames (nodes)
products = pd.DataFrame({
    'product_id': [1, 2, 3],
    'category': ['Electronics', 'Home', 'Other'],
    'cost': [300.0, 15.0, 500.0],
})

customers = pd.DataFrame({
    'customer_id': [100, 101, 102, 103],
    'customer_name': ['Acme Co', 'Globex', 'Initech', 'Umbrella'],  
    'segment': ['SMB', 'Enterprise', 'SMB', 'Consumer'],
    'region_id': [7, 8, 7, 9],  
})

regions = pd.DataFrame({
    'region_id': [7, 8, 9],
    'region': ['North', 'West', 'South'],
})

calendar = pd.DataFrame({
    'date': ['2024-01-01', '2024-01-02', '2024-01-03', '2024-01-04', '2024-01-05'],  
    'month': ['2024-01', '2024-01', '2024-01', '2024-01', '2024-01'],
})

sales = pd.DataFrame({
    'sale_id': [10, 11, 12, 13, 14, 15],
    'product_id': [1, 1, 2, 3, 2, 1],                        
    'customer_id': [100, 101, 102, 103, 100, 102],           
    'date': ['2024-01-01', '2024-01-02', '2024-01-03', '2024-01-04', '2024-01-05', '2024-01-03'],  
    'promo_code': ['NEW10', 'NONE', 'DISC5', 'NONE', 'DISC5', 'NEW10'],  
    'qty': [2, 1, 4, 3, 5, 2],
    'price': [500.0, 500.0, 25.0, 800.0, 25.0, 500.0],
})

promos = pd.DataFrame({
    'promo_code': ['NEW10', 'DISC5', 'NONE'],
    'promo_type': ['Launch', 'Discount', 'No Promo'],
})

# 2) Build the hypercube
cube = Hypercube({
    'Product': products,
    'Customer': customers,
    'Region': regions,
    'Calendar': calendar,
    'Sales': sales,
    'Promos': promos,
})

# Inspect your new hypercube model (shared columns are now the edges)

cube.visualize_graph(w=12, h=10, full_column_names=False)

# 3) Define metrics
cube.define_metric(
    name='Revenue',
    expression='[qty] * [price]',
    aggregation='sum'
)

cube.define_metric(
    name='Units',
    expression='[qty]',
    aggregation='sum'
)

cube.define_metric(
    name='Margin',
    expression='([price] - [cost]) * [qty]',
    aggregation='sum'
)

cube.define_metric(
    name='Number of Sales',
    expression='[sale_id]',
    aggregation='count'
)

# 4) Define query/ies
cube.define_query(
    query_name="sales_analysis",
    dimensions={'region', 'category', 'promo_type'},
    metrics=['Revenue', 'Units', 'Margin', 'Number of Sales']
)

# 5) Execute the query (or queries)
cube.query("sales_analysis")

Output:

   region     category promo_type  Revenue  Units  Margin  Number of Sales
0  North  Electronics     Launch   2000.0      4   800.0                2
1  North         Home   Discount    225.0      9    90.0                2
2  South        Other   No Promo   2400.0      3   900.0                1
3   West  Electronics   No Promo    500.0      1   200.0                1

# 6) Apply a filter and query again
cube.filter({'customer_name': ['Initech']})     
cube.query("sales_analysis")

Output:

   region     category promo_type  Revenue  Units  Margin  Number of Sales
0  North  Electronics     Launch   1000.0      2   400.0                1
1  North         Home   Discount    100.0      4    40.0                1

Notes

• Multi-hop works out of the box through implicit relationships (shared column names connect DataFrames, just make sure to rename your columns accordigly).

• If input tables are connected by more than one column, composite keys and tables will be created automatically. The original tables' columns will be replaced with the composite keys.

• The relationships graph can not contain cycles. The hypercube will not load if there are.

• Cardinallity is not assumed nor cheched (any-to-any). Links are treated as many-to-many, which can duplicate rows and skew aggregations if not modeled carefully.

• Columns in Metric Expression use square brackets: [qty], [price], [cost], etc.

• Metric Aggregation accepts pandas group by pre-defined aggregations (e.g., 'sum', 'mean') or a custom function callable.

• Addional features such as Context Filter Control, New Context States and Metric filters and row conditions are available but omitted here for brevity and simplicity. See the docs for details.

Full documentation

For concepts, API specs, advanced filtering, context states, full examples and Streamlit integration see:

• Docs: https://cube-alchemy.info

Creator

Created with 🧠 and ☕ by Juan C. Del Monte