omni-data-refinement 0.3.0

Omni Data Refinement — The standard framework for dataset quality, validation, and monitoring.

OMR — Omni Data Refinement

OMR is a pure Python framework for dataset quality, validation, profiling, monitoring, and reliability.

The goal is to become the standard tool developers use immediately after loading a dataset, similar to how Pandas is used for data manipulation.

OMR is designed to be Pure Python, Open Source, Modular, Extensible, Enterprise-ready, and entirely independent of external AI APIs or LLMs.

Core Vision

Instead of writing dozens of exploratory scripts, you just pass your dataset to OMR.

Current workflow:

import pandas as pd
df = pd.read_csv("data.csv")
# Write 50 lines of .isnull(), .duplicated(), and data exploration code...

Desired workflow:

import pandas as pd
from omr import Dataset

df = pd.read_csv("data.csv")
report = Dataset(df).health()

print(report.score)  # e.g. 87

---

Capabilities

OMR handles 12 complete domains of data intelligence:

1. Health Engine: 5-pillar quality score (Completeness, Uniqueness, Consistency, Validity, Conformity).
2. Cleaning Engine: Auto-resolution of missing values, duplicates, and type mismatches.
3. Profiling Engine: Deep statistical profile of every column.
4. Validation Engine: Schema-based validation with typed constraints (PositiveInteger, Email, etc).
5. Statistical Engine: Outlier detection, multicollinearity, skewness, class imbalance.
6. Drift Engine: Distribution shift detection using PSI, KS Test, and JS Divergence.
7. Monitoring System: Continuous tracking and alerting for data decay.
8. Explainability: Rule-based explanations for data science issues.
9. Versioning: Snapshots and rollback capabilities for transformations.
10. Reporting: Export to HTML, Markdown, or JSON.
11. Pipelines: Fluent, chainable API for applying operations.
12. Plugin Registry: Extensible via third-party domain-specific packages.

---

Installation

pip install omr

Dependencies: pandas, numpy, rich. (Optional: scipy for KS Drift tests). Compatible with Pandas, Polars, and NumPy.

---

Quickstart

1. Diagnostic Health Check

Get an immediate quality score and detailed issue list.

import pandas as pd
from omr import Dataset

df = pd.read_csv("messy_data.csv")
dataset = Dataset(df)

# Run health check
report = dataset.health()
print(f"Health Score: {report.score}/100")

2. Auto-Cleaning

Automatically fix missing values, duplicate rows, and mixed data types.

dataset.clean()
dataset.explain_changes() # View transformation log

3. Schema Validation

Enforce strict business rules on your dataset.

from omr import schemas

schema = {
    "age": schemas.PositiveInteger(max=120),
    "salary": schemas.PositiveFloat(min=10000),
    "status": schemas.OneOf("active", "inactive"),
    "email": schemas.Email()
}
dataset.validate(schema)

4. Advanced Analytics & Drift

Detect complex issues or compare against production data.

# Detect outliers, multicollinearity, and zero variance features
dataset.analyze()

# Compare against yesterday's data to detect drift
prod_dataset = Dataset(pd.read_csv("prod_data.csv"))
dataset.compare(prod_dataset)

5. Fluent API (Pipelines)

Chain commands together.

clean_df = (Dataset(df)
    .health()
    .clean()
    .analyze()
    .export())

---

Architecture (SOLID Principles)

OMR is modular and testable. The architecture is cleanly divided into:

• omr.core: The main Dataset interface and DatasetMetadata state tracking.
• omr.health, omr.cleaning, omr.profiling, omr.validation, omr.statistics, omr.drift, omr.monitoring, omr.explainability, omr.versioning, omr.pipelines, omr.reports, omr.plugins, omr.integrations, omr.utils, omr.schemas.

---

Extensibility (Plugins)

OMR supports domain-specific extensions.

from omr import register_plugin

@register_plugin("medical")
class MedicalPlugin:
    def validate_ehr(self, df):
        pass

---

License

MIT License. See LICENSE for details.