Feature Highlights

Screenshots

Load — bundled datasets, a file, or a URL. Dates are recognised on the way in, and the shape and memory are reported before you commit to anything.	Pivot — counts, means and percentages, with margins. Every table has a Show Code panel underneath carrying the pandas that produced it.
Analyze — the result is a sentence, and it says which level it treated as a success, because that choice changes the answer.	Evaluate — accuracy next to the majority-class baseline it has to beat, and a threshold that moves precision against recall.

Quick Start

Launch the interactive workbench

pyanalytica                # CLI entry point (after pip install)
python -m pyanalytica      # or run as a module

Use as a Python library

Every analytics function returns a (result, CodeSnippet) tuple. The CodeSnippet contains the equivalent pandas/sklearn code so students can see what runs under the hood.

from pyanalytica.data.load import load_bundled
from pyanalytica.data.profile import profile_dataframe
from pyanalytica.visualize.distribute import histogram
from pyanalytica.visualize.relate import scatter
from pyanalytica.explore.summarize import group_summarize

# Load a bundled dataset
df, code = load_bundled("tips")

# Profile the dataframe — column types, missing values, summary stats
profile = profile_dataframe(df)

# Visualize
fig, code = histogram(df, "total_bill", bins=20)
fig, code = scatter(df, x="total_bill", y="tip", color_by="smoker")

# Summarize — group_cols, value_cols, agg_funcs are all lists
result, code = group_summarize(
    df,
    group_cols=["day"],
    value_cols=["tip"],
    agg_funcs=["mean"],
)

The CodeSnippet Pattern

Every analytics function in PyAnalytica returns a tuple of (result, CodeSnippet). The CodeSnippet dataclass holds the equivalent pandas/sklearn code so students can learn what happens behind the UI:

from pyanalytica.core.codegen import CodeSnippet

# CodeSnippet(code="df.groupby(['day'])['tip'].mean()", imports=["import pandas as pd"])

# In the Shiny UI, the "Show Code" button renders this as a copyable code block.
# The emitted code uses real pandas/sklearn calls — never wrapper functions.

Installation

# Core package (Shiny UI + all analytics)
pip install pyanalytica

# With AI integration (Anthropic Claude)
pip install "pyanalytica[ai]"

# With Jupyter notebook export
pip install "pyanalytica[report]"

# Everything (recommended)
pip install "pyanalytica[all]"

To update to the latest version:

pip install --upgrade pyanalytica

Switching from a GitHub install? Run pip uninstall pyanalytica first, then install from PyPI above.

Install from source (for development)

git clone https://github.com/AshishK714/PyAnalytica.git
cd PyAnalytica
pip install -e ".[dev,all]"

Bundled Datasets

from pyanalytica.datasets import list_datasets, load_dataset

list_datasets()          # ['candidates', 'companies', 'diamonds', 'events', 'jobs', 'tips']
df = load_dataset("diamonds")

To regenerate bundled datasets:

PYTHONPATH=src python -m pyanalytica.datasets.generate

Architecture Overview

┌─────────────────────────────────────────────────────────┐
│                    Shiny for Python UI                   │
│  ┌──────────────────────────────────────────────────┐   │
│  │  Modules: mod_load, mod_profile, mod_view, ...   │   │
│  └──────────────┬───────────────────────────────────┘   │
│                 │                                        │
│  ┌──────────────▼───────────────────────────────────┐   │
│  │  Components: dataset_selector, code_panel,       │   │
│  │  decimals_control, chat_panel, download_result   │   │
│  └──────────────┬───────────────────────────────────┘   │
├─────────────────┼───────────────────────────────────────┤
│                 │      Analytics Packages                │
│  ┌──────────────▼───────────────────────────────────┐   │
│  │  data/   explore/   visualize/   analyze/        │   │
│  │  model/  homework/  report/      ai/             │   │
│  └──────────────┬───────────────────────────────────┘   │
│                 │                                        │
│  ┌──────────────▼───────────────────────────────────┐   │
│  │  Core: codegen, state, config, theme, profile,   │   │
│  │  model_store, procedure, session, column_utils   │   │
│  └──────────────────────────────────────────────────┘   │
└─────────────────────────────────────────────────────────┘

The architecture follows a package-first design:

• Core provides shared utilities (CodeSnippet generation, state management, configuration)
• Analytics packages (data/, explore/, visualize/, analyze/, model/) contain pure functions that work independently of any UI
• UI modules in ui/modules/ call analytics functions and handle Shiny reactivity
• WorkbenchState is a simple data store; the Shiny reactive graph manages the current selection

Configuration

User Profile

PyAnalytica reads user preferences from ~/.pyanalytica/profile.yaml (auto-created on first use):

# ~/.pyanalytica/profile.yaml
api_key: ""          # Anthropic API key for AI features
decimals: 3          # Default decimal places for numeric output
theme: default       # UI theme

# Instructor fields (optional)
instructor_name: ""
institution: ""
course: ""

Precedence: Environment variable > profile.yaml > built-in default

Course Configuration

Instructors can place a pyanalytica.yaml in the working directory to control which menu items are visible (with optional date-gating):

menus:
  - name: Data
    visible: true
  - name: Model
    visible: true
    after: "2025-02-15"   # Only show after this date
  - name: Homework
    visible: true

# homework1.yaml
title: "Homework 1: Exploratory Data Analysis"
dataset: tips
due_date: "2025-03-01"
questions:
  - id: q1
    type: numeric
    prompt: "What is the mean total bill?"
    answer_hash: "sha256:..."    # Hash of the correct answer
    tolerance: 0.01
  - id: q2
    type: multiple_choice
    prompt: "Which day has the highest average tip?"
    choices: ["Thur", "Fri", "Sat", "Sun"]
    answer_hash: "sha256:..."
  - id: q3
    type: dataframe
    prompt: "Create a summary table of mean tip by day"
    answer_hash: "sha256:..."

from pyanalytica.homework.schema import hash_answer
hash_answer(19.7859)    # 'sha256:...'
hash_answer("Sun")      # 'sha256:...'

AI Features

PyAnalytica includes four AI-powered modules that work in rule-based mode by default and can be enhanced with an Anthropic API key:

Set your API key via environment variable or user profile:

export ANTHROPIC_API_KEY="sk-ant-..."

Procedure Builder & Reports

Recording workflows

The Procedure Builder records every analytics operation as a reproducible step:

1. Click Start Recording in the Report > Procedure tab
2. Perform your analysis (load data, transform, visualize, model, etc.)
3. Each step is captured with its code snippet and can be annotated with comments
4. Stop Recording when done

Export formats

from pyanalytica.core.procedure import Procedure

proc = Procedure.from_json("my_analysis.json")
proc.to_python("my_analysis.py")
proc.to_notebook("my_analysis.ipynb")

Development

Setup

git clone https://github.com/AshishK714/PyAnalytica.git
cd PyAnalytica
pip install -e ".[dev,all]"

# Generate bundled datasets
PYTHONPATH=src python -m pyanalytica.datasets.generate

Run tests

PYTHONPATH=src python -m pytest tests/ -v

Build

pip install build
python -m build

Project structure

PyAnalytica/
├── src/pyanalytica/
│   ├── __init__.py              # Package version
│   ├── __main__.py              # python -m pyanalytica entry
│   ├── core/                    # Shared utilities
│   │   ├── codegen.py           # CodeSnippet + on_record hook
│   │   ├── column_utils.py      # ColumnType classification
│   │   ├── config.py            # CourseConfig + menu visibility
│   │   ├── model_store.py       # ModelArtifact + ModelStore
│   │   ├── procedure.py         # ProcedureStep / Procedure / Recorder
│   │   ├── profile.py           # UserProfile + get_api_key()
│   │   ├── session.py           # Session save / load / list
│   │   ├── state.py             # WorkbenchState
│   │   └── theme.py             # Theme management
│   ├── data/                    # Load, profile, transform, combine, export
│   ├── ask/                     # describe_one, relate_two: the column types pick the method
│   ├── explore/                 # Summarize, pivot, crosstab
│   ├── visualize/               # Distribute, relate, compare, correlate, timeline
│   ├── analyze/                 # Means, proportions, correlation
│   ├── model/                   # Regression, classify, cluster, reduce, evaluate, predict
│   ├── homework/                # Schema, loader, grader, submission
│   ├── report/                  # Notebook + export
│   ├── ai/                      # Interpret, suggest, challenge, query
│   ├── datasets/                # Bundled CSV data + generator
│   └── ui/                      # Shiny application
│       ├── app.py               # Main app entry point
│       ├── www/style.css         # Glassmorphism CSS theme
│       ├── components/          # Reusable UI components
│       └── modules/             # Feature modules (data/, ask/, explore/, visualize/, ...)
├── tests/                       # 274 tests across 42 test files
├── pyproject.toml               # Build config (hatchling)
├── CHANGELOG.md                 # Version history
└── LICENSE                      # MIT License

Contributing

1. Fork the repository
2. Create a branch for your feature (git checkout -b feature/my-feature)
3. Write tests for new functionality
4. Run the test suite to ensure all tests pass
5. Submit a pull request with a clear description

Code style

• All analytics functions return (result, CodeSnippet) tuples
• CodeSnippets emit real pandas/sklearn code, never wrapper calls
• Use ColumnType for column classification instead of ad-hoc dtype checks
• Keep UI modules thin — business logic belongs in analytics packages

License

MIT License. Copyright 2026 Ashish Khandelwal.

See LICENSE for details.

Acknowledgements

PyAnalytica is inspired by Radiant by Vincent Nijs (UC San Diego) — a comprehensive R/Shiny analytics platform for business education.

Built with Shiny for Python, pandas, scikit-learn, matplotlib, seaborn, and SciPy.

AI features powered by Anthropic Claude.

Developed for teaching at the University of Illinois at Urbana-Champaign.