EazyDataFix

Deterministic data transformation you can inspect and trust

EazyDataFix is a deterministic-first Python library for turning messy datasets into reliable, typed, auditable data for analysis and downstream tools.

It combines dataset profiling, quality assessment, controlled cleaning, safe type conversion, preparation, contracts, and before-and-after reporting. The active roadmap is focused on Analysis Ready, ML Ready, and Power BI Ready data profiles. Existing EDA and Agentic EDA APIs remain available for compatibility, but are not the current development focus.

EazyDataFix v1.4.1 is the current stable production release. It preserves the v1 API and adds deterministic Analysis Ready, leakage-safe ML Ready, and validated Power BI Ready workflows.

Install with pip install eazydatafix · Documentation · PyPI

Quick start

Preview a controlled transformation without modifying the source dataset:

import eazydatafix as edf

preview = edf.fix(
    "employees.csv",
    edf.FixConfig(
        dry_run=True,
        numeric_conversion_threshold=0.95,
        date_parsing_threshold=0.80,
    ),
)

print(preview.proposed_dataset)
print(preview.change_log)

Apply the same rules and continue with the prepared DataFrame:

result = edf.fix(
    "employees.csv",
    edf.FixConfig(
        numeric_conversion_threshold=0.95,
        date_parsing_threshold=0.80,
    ),
)
preparation = edf.prepare_with_report(result.dataset)

df = preparation.dataset
print(preparation.changes)
print(df.dtypes)

Or use the existing analysis-ready convenience workflow:

df = edf.analysis_ready("employees.csv")

Return the same prepared dataset with readiness evidence:

result = edf.analysis_ready_with_report(
    "employees.csv",
    edf.AnalysisReadyConfig(
        prepare_config=edf.PrepareConfig(
            outlier_action="flag",
            derive_date_parts=("year", "month", "day_of_week"),
        ),
    ),
)

print(result.before_score, result.after_score, result.is_ready)
print(result.changes)
print(result.warnings)
print(result.validations)

df = result.dataset

The detailed workflow composes assessment, controlled fixing, preparation, and final validation. It also reports constant, nearly empty, invalid, and inconsistently labelled category columns without hiding them from the user.

Prepare leakage-safe supervised machine-learning inputs:

result = edf.ml_ready(
    "customer_churn.csv",
    target="churned",
    config=edf.MLReadyConfig(
        test_size=0.20,
        categorical_encoding="one_hot",
        scaling="standard",
    ),
)

X_train, X_test = result.X_train, result.X_test
y_train, y_test = result.y_train, result.y_test

print(result.readiness_score, result.is_ready)
print(result.feature_names)
print(result.issues)

# Reuse parameters fitted only from X_train.
result.artifact.to_json("churn_preprocessing.json")

ml_ready(...) requires an explicit target, splits rows before fitting learned transformations, and never trains or evaluates a model. Identifiers, constants, high-cardinality features, unsupported datetimes, and likely leakage fields are reported and safely excluded by default. Missing target rows are dropped before splitting; the target itself is never imputed, encoded, or scaled.

Prepare a validated multi-table Power BI model input:

tables = {
    "sales": "sales.csv",
    "customers": "customers.csv",
}

result = edf.powerbi_ready(
    tables,
    edf.PowerBIReadyConfig(
        keys=(edf.PowerBIKey("customers", ("customer_id",)),),
        relationships=(
            edf.PowerBIRelationship(
                "sales",
                ("customer_id",),
                "customers",
                ("customer_id",),
                "many_to_one",
            ),
        ),
        generate_date_table=True,
        date_columns={"sales": ("order_date",)},
    ),
)

print(result.readiness_score, result.is_ready)
print(result.field_types)
print(result.issues)

result.save("powerbi_output", formats=("csv", "excel"))

powerbi_ready(...) also accepts a single DataFrame or file. It composes the Analysis Ready workflow, normalizes collision-free names and supported field types, flattens record-valued fields, validates keys and relationship cardinality, and can generate a continuous date dimension. Exports include a JSON readiness report; Parquet uses the existing parquet extra. EazyDataFix prepares model inputs but intentionally does not generate .pbix reports, visuals, measures, or dashboards.

The controlled transformation pipeline can:

1. Normalize column names and text whitespace
2. Detect configured missing-value markers
3. Convert safe numeric, currency, percentage, boolean, and date values
4. Preserve identifiers, phone numbers, emails, and leading zeroes
5. Remove configured duplicates and empty structures
6. Record every applied or proposed change

Installation

pip install eazydatafix

For Parquet support:

pip install "eazydatafix[parquet]"

For optional scikit-learn transformer interoperability:

pip install "eazydatafix[ml]"

Requires Python 3.10 or later. Tested with Python 3.10–3.13.

Why EazyDataFix

Deterministic First

Metrics, findings, and recommendations come from reproducible calculations.

Traceable Decisions

Plans, actions, questions, and visualisations identify their source analysis step.

Safe by Default

Caller DataFrames are not mutated by the deterministic EDA workflow.

AI Optional

The deterministic workflow does not require an LLM. The v0.5.0 release adds optional grounded narratives through a provider adapter; existing workflows continue to run without an API key or AI dependency.

Optional grounded AI narrative

Create a business-facing narrative only after deterministic analysis is complete. The provider receives an immutable, compact evidence brief, not the raw dataset. Every generated statement must cite one or more evidence IDs from that brief. EazyDataFix rejects malformed or unknown citations, invented numbers, unsupported causal language, and claims without sufficient lexical support in their cited evidence. The narrative is bound to the exact workflow by a SHA-256 fingerprint, so it cannot be exported with a different or modified workflow.

These deterministic checks reduce unsupported output but cannot prove the semantic truth of AI-written text. Review the narrative before using it for a decision. HTML and Markdown reports include an evidence-reference section for that review.

import eazydatafix as edf
from eazydatafix.narratives import OpenAINarrativeProvider

workflow = edf.run_agentic_eda("employees.csv")
provider = OpenAINarrativeProvider(model="your-openai-model")

narrative = edf.generate_agentic_eda_narrative(workflow, provider)

report = edf.export_agentic_eda_report(
    workflow,
    output_dir="eda-report",
    formats=["html", "json", "markdown"],
    narrative=narrative,
)

Install the adapter only when needed:

pip install "eazydatafix[openai]"

Workflow

flowchart LR
    A[Dataset] --> B[Understand]
    B --> C[Assign Semantic Roles]
    C --> D[Plan Analyses]
    D --> E[Execute]
    E --> F[Generate Findings and Actions]
    F --> G[Export Reports and Visualisations]

Current capabilities

Data Quality

• Missing-value analysis
• Duplicate detection
• Completeness checks
• Validity checks
• Consistency checks
• Accuracy checks
• Timeliness checks
• Data-quality scoring

Deterministic EDA

• Numeric analysis
• Categorical analysis
• Boolean analysis
• Datetime analysis
• Correlation review
• IQR outlier analysis
• Skewness analysis
• Class-imbalance analysis

Agentic Workflow

• Semantic column-role detection
• Deterministic analysis planning
• Modular analysis execution
• Priority findings
• Traceable follow-up actions
• Visualisation recommendations
• Unresolved domain questions
• Partial-failure isolation
• Human approval checkpoints between planning and execution
• Dataset fingerprint validation before approved execution

Reporting

• Console
• HTML
• PDF
• Excel
• CSV
• JSON
• Markdown
• Deterministic PNG visualisations
• Ready-to-run Jupyter Notebook export

Input Support

• pandas DataFrames
• CSV
• Excel
• JSON
• Parquet with the optional dependency

Example output

A data-quality assessment can produce a concise console summary:

EASYDATAFIX DATA QUALITY REPORT

Score         : 90.37 / 100
Grade         : A
Completeness  : 96.97%
Uniqueness    : 100.00%
Validity      : 55.00%
Consistency   : 100.00%
Accuracy      : 100.00%
Timeliness    : 100.00%

An Agentic EDA report with HTML, JSON, and optional Markdown output can produce:

eda-report/
├── agentic-eda-report.html
├── agentic-eda-report.json
├── agentic-eda-report.md
└── visualisations/
    ├── 01-missing-value-chart-phone-salary.png
    ├── 02-bar-chart-department.png
    └── 03-time-series-line-chart-joining-date.png

HTML and JSON are generated by default; Markdown is generated when requested. The exact charts depend on the dataset and the workflow's deterministic visualisation recommendations.

API overview

Detailed API documentation is maintained on the documentation website.

Resources

• Project Website
• Getting Started
• Documentation
• API Reference
• Roadmap
• Changelog
• PyPI
• GitHub Repository
• GitHub Issues

Project status

• Current stable version: v1.4.1
• Development status: Production/Stable
• Released: 26 August 2026
• Python support: 3.10–3.13
• Licence: MIT

The v1 public API follows Semantic Versioning. Backward-incompatible public API changes require a new major version.

Roadmap preview

• v0.3.0 — Deterministic Agentic EDA Foundation — Released
• v0.4.0 — Notebook Export and Human Approval — Released
• v0.5.0 — Optional Grounded AI Narratives — Released
• v0.6.0 — Controlled, Auditable Cleaning — Shipped in v1.0.0
• v0.7.0 — Data Preparation and Feature Readiness — Shipped in v1.0.0
• v0.8.0 — Data Validation and Contracts — Shipped in v1.0.0
• v0.9.0 — Production Workflow — Shipped in v1.0.0
• v1.0.0 — Stable Production API — Released
• v1.2.0 profile — Analysis Ready — Shipped in v1.4.0
• v1.3.0 profile — ML Ready — Shipped in v1.4.0
• v1.4.0 — Power BI Ready and transformation-first release — Shipped
• v1.4.1 — Reproducible packaging maintenance release — Current

See the full roadmap for milestone details.

Contributing

Contributions, issue reports, and focused feature proposals are welcome. See CONTRIBUTING.md for the development workflow, quality checks, and pull-request guidance. Use GitHub Issues to report bugs or discuss a focused change.

Licence

EazyDataFix is available under the MIT Licence.