murthylytics 0.1.0

Production-grade, AI-powered machine learning data-analysis and preparation library.

Murthylytics

Production-grade, AI-powered machine-learning data analysis & preparation for Python.

Murthylytics turns hundreds of lines of exploratory-data-analysis and preprocessing boilerplate into a handful of intuitive commands — while staying transparent, configurable and production-quality. Think of it as a next-generation companion to pandas, ydata-profiling, Sweetviz and scikit-learn's preprocessing tools, unified behind one stateful API.

import murthylytics as mlt

mlt.read_csv("data.csv")   # auto-detects encoding, delimiter, dtypes, datetimes

mlt.summary()              # compact overview
mlt.inspect()              # intelligent profiling: roles, targets, leakage, issues

mlt.clean_data()           # roadmap: intelligent cleaning & imputation
mlt.auto_eda()             # roadmap: HTML / Markdown / PDF / JSON reports
mlt.preprocess()           # roadmap: task detection + preprocessing pipeline
mlt.recommend_model()      # roadmap: model recommendations with justification

Why Murthylytics

• One stateful facade. Load a dataset once; every call operates on the active session — no handle passing.
• Intelligent by default. Encoding/delimiter/dtype detection, column-role inference, target detection, class-imbalance and data-leakage checks, all automatic.
• Lightweight core, optional power. Base install is just pandas + numpy. Heavy tooling (plots, XGBoost, SHAP, Parquet…) is lazy-loaded behind extras.
• Engineered to last. src/ layout, full type hints, SOLID modules, a pure stateless engine under the facade, and a test suite.

Installation

pip install murthylytics            # core (pandas + numpy)
pip install 'murthylytics[stats]'   # scipy + scikit-learn powered analysis
pip install 'murthylytics[viz]'     # matplotlib + seaborn plotting
pip install 'murthylytics[all]'     # everything

Develop locally: pip install -e '.[dev,stats]'

Quick start

import murthylytics as mlt

mlt.read_csv("titanic.csv")

mlt.shape()          # {'rows': 891, 'columns': 12}
mlt.missing()        # per-column missing counts & %
report = mlt.inspect()
print(report)                    # human-readable summary
print(report.recommendations)    # actionable next steps
report.to_dict()                 # JSON-serialisable

Feature status

Area	API	Status
Data I/O (CSV/Excel/JSON/Parquet/SQL)	read_*	✅ available
Exploration	show/top/bottom/shape/columns/summary/describe/info/types/missing/duplicates/memory	✅ available
Intelligent inspection	inspect	✅ available
Intelligent cleaning	clean_data	🚧 roadmap (iteration 2)
Visualization	cleaned_pyplot	🚧 roadmap (iteration 3)
AutoEDA reports	auto_eda	🚧 roadmap (iteration 4)
Preprocessing	preprocess	🚧 roadmap (iteration 5)
Feature engineering / selection	engineer_features / select_features	🚧 roadmap (iteration 6)
Model recommendation	recommend_model	🚧 roadmap (iteration 7)
Pipeline export	export_pipeline	🚧 roadmap (iteration 8)
Plugins (malware, finance, …)	use_plugin	🚧 roadmap (iteration 9)

The roadmap functions are already part of the public surface with stable signatures; calling one today raises a clear NotImplementedError pointing at the iteration that delivers it.

Architecture

murthylytics/
├── core/          config · session/context · logging · exceptions · types
├── io/            smart readers · encoding/delimiter/dtype detection · memory opt
├── exploration/   display helpers · intelligent inspection
├── cleaning/      (roadmap) duplicates · imputation · outliers · text fixes
├── viz/           (roadmap) auto-plot dispatcher
├── eda/           (roadmap) HTML/MD/PDF/JSON reports
├── preprocess/    (roadmap) task detection · scaling · encoding · pipelines
├── features/      (roadmap) engineering · selection
├── modeling/      (roadmap) model recommendation
├── pipeline/      (roadmap) save/load/export
└── plugins/       (roadmap) domain plugins

All mutable state lives in a single Session; the engine modules are pure functions, which keeps them independently testable and reusable.

License

MIT © Pranav Murthy