dataprism 0.1.6

A Python library for exploratory data analysis with advanced statistical features

DataPrism

A Python library for exploratory data analysis with data profiling, quality assessment, and stability monitoring.

Interactive Viewer

DataPrism includes a built-in interactive dashboard to explore your analysis results in the browser.

from dataprism import DataPrism, DataLoader

# Load data from CSV or Parquet
df = DataLoader.load_csv("data.csv")
# df = DataLoader.load_parquet("data.parquet")

# Run analysis and launch viewer
prism = DataPrism()
prism.analyze(
    data=df,
    target_variable="target",
    exclude_columns=["id", "split", "onboarding_date"],
    output_path="eda_results.json",
)
prism.view()

Summary — Dataset overview, insights, top features by IV, data quality score, and provider match rates.

Catalog — Sortable feature table with type, provider, target correlation, IV, and PSI at a glance.

Deep Dive — Per-feature detail view with statistics, violin plots, distribution charts, PSI trend analysis, target associations, and correlations.

Associations — Mixed-method heatmap (Pearson, Theil's U, Eta) showing relationships across all features.

How DataPrism Compares

Capability	DataPrism	ydata-profiling	Sweetviz	D-Tale	AutoViz	DataPrep
Predictive power (IV / WoE)	✅	➖	🟡	🟡	➖	➖
Drift detection (PSI)	✅	🟡	🟡	➖	➖	🟡
Data quality score	✅	➖	➖	➖	➖	➖
Multi-source match rates	✅	➖	➖	➖	➖	➖
Schema-aware profiling	✅	🟡	🟡	🟡	➖	🟡
Structured JSON output	✅	✅	➖	🟡	➖	🟡
Interactive explorer	✅	✅	🟡	✅	🟡	✅

✅ Supported 🟡 Partial ➖ Not supported

Installation

pip install dataprism

Quick Start

from dataprism import DataPrism, DataLoader

df = DataLoader.load_csv("data.csv")

prism = DataPrism()
results = prism.analyze(
    data=df,
    exclude_columns=["customer_id", "created_at"],
    target_variable="target",
    output_path="eda_results.json"
)

For schema-aware profiling, stability analysis, and advanced configuration, see the Usage Guide.

Roadmap

DataPrism is being built for the AI era — where data analysis is increasingly driven by LLM agents, automated pipelines, and programmatic consumers rather than humans clicking through dashboards.

AI-Native Analysis

• Natural language insights — Auto-generated plain-English summaries of each feature, anomalies, and recommendations that LLMs can directly incorporate into reports.

Closing the Gaps

• Dataset comparison — Side-by-side train/test/production profiling with automatic drift highlights.
• Scatter & pair plots — Interactive scatter matrices for continuous feature pairs with target coloring.
• Auto-visualization — One-line generation of per-feature visual summaries exportable as images.
• Spark/Dask support — Distributed computation for datasets that don't fit in memory.
• Streaming analysis — Incremental profiling for real-time data pipelines without re-analyzing the full dataset.

Deeper Intelligence

• Automated feature recommendations — Go beyond flagging issues to suggesting transformations (log, binning, encoding) based on distribution shape and target relationship.
• Anomaly explanations — When outliers or drift are detected, surface the likely cause (data pipeline issues, population shift, seasonality).
• Cross-dataset lineage — Track how feature distributions evolve across model versions and data refreshes.

Documentation

• Usage Guide — schema, stability analysis, advanced configuration, provider match rates
• Architecture — internals, module structure, data flow
• Decision Records — key design decisions and rationale
• Examples — usage examples and demos

Development

pip install -e .           # Install for development
python -m build            # Build package
python -m pytest tests/    # Run tests

Requirements

• Python 3.9+
• pandas >= 2.0.0
• numpy >= 1.24.0
• scipy >= 1.10.0
• pyarrow >= 10.0.0 (for Parquet support)

License

MIT License - see LICENSE file for details.

Contact

For questions or suggestions:

• Email: [e-mail]
• GitHub: https://github.com/lattiq/dataprism

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.