EazyDataFix 1.0.0

One stable workflow

edf.run() composes the four stages most data projects repeat: profiling, quality assessment, controlled cleaning and deterministic EDA. The returned RunResult keeps each stage separate, so applications can inspect exactly what happened.

1import eazydatafix as edf23result = edf.run("employees.csv")45print(result.profile)6print(result.assessment.quality.score)7print(result.fix_result.applied_fixes)8print(result.eda_result.observations)

Controlled, auditable cleaning

Cleaning is no longer an all-or-nothing operation. A dry run keeps the source dataset unchanged, exposes the proposed dataset separately and records structured before/after changes. Column rules override the dataset-wide strategy only where needed.

1import eazydatafix as edf23config = edf.FixConfig(4    dry_run=True,5    column_rules={6        "salary": edf.ColumnCleaningRule(7            missing_value_strategy="mean"8        ),9        "notes": edf.ColumnCleaningRule(10            trim_whitespace=False11        ),12    },13)1415preview = edf.fix("employees.csv", config)16print(preview.change_log)17print(preview.proposed_dataset.head())

Preparation with evidence

edf.prepare_with_report() returns the prepared DataFrame together with applied changes, warnings, before/after shapes and data types. Thresholds control numeric and date conversion, while outliers can be left unchanged, capped or dropped.

1import eazydatafix as edf23config = edf.PrepareConfig(4    numeric_conversion_threshold=0.95,5    date_parsing_threshold=0.80,6    outlier_action="cap",7    normalize_text=True,8)910report = edf.prepare_with_report("employees.csv", config)11print(report.changes)12print(report.warnings)13prepared_df = report.dataset

Data contracts and quality rules

Infer a contract from a trusted dataset, then validate future inputs against expected fields and types. Add explicit not_null, unique, minand max rules for pipeline decisions.

1import eazydatafix as edf23contract = edf.infer_schema("baseline.csv")4rules = (5    edf.QualityRule("id_unique", "employee_id", "unique"),6    edf.QualityRule("salary_non_negative", "salary", "min", 0),7)89report = edf.validate_contract("incoming.csv", contract, rules)10print(report.passed)11print(report.to_dict())

Production command-line workflows

The edf command runs profile, assess, fix, prepare, EDA, Agentic EDA or the complete workflow. It accepts direct file paths or JSON/YAML configuration, writes JSON summaries and JSONL events, and returns exit code 0, 1 or 2 for success, processing failure or configuration failure.

1edf employees.csv23edf sales.csv customers.csv \4  --config workflow.yaml \5  --output batch-results.json \6  --log-file workflow-events.jsonl

Compatibility and migration

No migration is required for existing v0.5 calls. Profiling, assessment, cleaning, preparation, deterministic and Agentic EDA, reports, approval checkpoints, notebook export and grounded narratives remain available. Applications can now catch the stable package-level EazyDataFixError; the historical spelling remains available for compatibility.

Install

pip install eazydatafix==1.0.0