Pranaam documentation

Pranaam returns calibrated estimates of whether a name follows Muslim- or non-Muslim-associated patterns in its training data. The result is a name-pattern estimate, not a statement about a person’s identity.

• Installation
  • Streamlit app
  • Model cache and offline use
  • Metadata schemas
• Quick start
• API Reference
  • Main Functions
  • Core Classes
  • Utility Functions
  • Logging Configuration
• Examples
  • Add estimates to a DataFrame
  • Process a CSV file
• Model v3 evidence audit
  • Claims and estimands
  • Paired comparison
  • Data integrity and support
  • Audit findings
  • Downstream packages
  • Check matrix
  • Rejected claims
  • Plan

Basic use

from pranaam import estimate_muslim_name_pattern

result = estimate_muslim_name_pattern(["Shah Rukh Khan", "Amitabh Bachchan"], lang="eng")
print(result)

The result contains a calibrated 0-to-1 score, an associated pattern or abstention, explicit script and UTF-8 byte-limit support, the score’s reference and calibration populations, its label source, and immutable model provenance. Pranaam downloads checksum-verified PyTorch model files from a pinned Hugging Face release on first use and stores them in the Hugging Face cache.

hindi = estimate_muslim_name_pattern(["शाहरुख खान", "अमिताभ बच्चन"], lang="hin")

Use these estimates only for aggregate research with suitable validation. Do not use them to label individuals or make consequential decisions.

Indices

• Index

• Module Index

• Search Page