mlplt 1.2.6

The complete ML toolkit — EDA, cleaning, training, explainability, deployment

mlpilot 🚀

The complete, unbreakable machine learning toolkit — from raw data to production in seconds.

---

🌟 The Vision: "Unbreakable Data Science"

Most Machine Learning libraries fail when things get messy. mlpilot is the first library built to be Unbreakable. It is designed for the real world—where data has typos, models hallucinate, and servers have legacy encodings.

One Import. Every Tool.
Whether you are a student in Google Colab, a Data Scientist on a local laptop, or a Production Engineer deploying to Docker, mlpilot provides a self-healing, zero-config environment that just works.

• Self-Healing 2.0: Real-time syntax correction, "Hallucination Immunity," and now Categorical Healing for raw string features.
• Global Encoding Shield: Revolutionary protection against UnicodeEncodeError on legacy Windows terminals (CP1252/ASCII).
• Silent Production: Absolute suppression of 3rd-party technical noise and clutter.

---

⚡ Quick Start: The "Omni-Pipeline"

Execute a professional, industrial ML workflow in 10 lines of code:

import mlpilot as ml
import seaborn as sns

# 1. Load data
df = sns.load_dataset('titanic')

# 2. The Unbreakable Pipeline
clean = ml.clean(df, target='survived')              # Hardens data against noise
X, y  = ml.features(clean.df, target='survived')     # Zero-leakage engineering
blitz = ml.baseline(X, y)                            # 10s "Winner-Takes-All" model search
tuned = ml.tune(blitz.best_name, X, y, time_budget=30) # Smart hyperparameter tuning
audit = ml.audit(tuned.best_model, X, y)             # Fairness & Bias verification

# 3. AI Insights
ml.analyst(df).ask("What is the survival rate for females?", auto_run=True)

---

📋 The Clear Vision: Why mlpilot?

Challenge	The Hard Way (Manual)	The mlpilot Way	Real Impact
Data Cleaning	50+ lines of loops, imputers, and drops.	ml.clean(df)	98% less code. No logic errors.
EDA Insights	100+ lines of matplotlib and seaborn.	ml.analyze(df)	Professional reports in 5 seconds.
Model Search	Manually trying 10+ algorithms in a loop.	ml.baseline(X, y)	Finds the winner in 10 seconds.
Deep Tuning	Complex GridSearchCV or Optuna setup.	ml.tune(name, X, y)	Best hyperparams with zero math.
Deployment	200+ lines (FastAPI, Docker, Pickling).	ml.deploy(model)	Production-ready API in 1 line.

---

🛠️ The Module Encyclopedia

🧬 Phase 1: Data Foundations & Cleansing

Master your raw data before it hits the model.

• ml.analyze(df): Generates a 12-section SmartEDA report with automatic leakage detection and quality scoring.
• ml.clean(df): The ultimate hardening tool. Automatically handles missing values, outliers, duplicates, and inconsistent categorical dtypes.
• ml.validate(df, schema): Industrial-grade data validation. Refreshes and verifies your data against a strictly inferred schema.

🧪 Phase 2: The Predictive Core

Turn raw tables into high-performance predictions.

• ml.features(df): Zero-leakage automated feature engineering. Handles target encoding, scaling, and rolling aggregations.
• ml.baseline(X, y): A high-speed model search tournament. Compares 12+ model families to find the best baseline.
• ml.tune(name, X, y): Smart, budget-aware hyperparameter tuning using a local optimization engine.
• ml.evaluate(model, X, y): Generates a 5-metric technical report with confusion matrices and performance charts.

🔍 Phase 3: specialize & specialized Insight

Beyond simple accuracy—understand and improve your models.

• ml.explain(model, X): Professional local/global interpretability (SHAP). Tells you exactly why a prediction was made.
• ml.balance(df): Fixes class imbalance using intelligent oversampling (SMOTE) and undersampling techniques.
• ml.forecast(series): TimeSense — Advanced multi-model time-series forecasting for finance and trends.

🎭 Phase 4: NLP & Text Intelligence

State-of-the-art Natural Language processing with zero training required.

• ml.sentiment(series): Zero-shot emotion and sentiment extraction from raw text.
• ml.topics(series): Automatically extracts core topics and themes from thousands of comments.
• ml.embed(series): Generates state-of-the-art vector embeddings for semantic search and clustering.

🤖 Phase 5: AI, Audit & Production

The future of Data Science—automated, ethical, and deployable.

• AI Analyst: Natural Language Interface. Ask complex questions about your data in plain English.
• MLAudit: Comprehensive bias and fairness auditing. Generates automated Model Cards to ensure your AI is ethical.
• Data Story: Merges EDA and Performance results into a professional narrative report for executives.
• LaunchPad: Generates a production-ready FastAPI application and Dockerfile for any model in one command.
• Session: The ultimate reproducibility tool. Records your entire workflow as a clean .py or .ipynb script.

---

🎯 Real-World Scenarios: Where mlpilot Shines

1. "The Messy CSV": You have a dataset with typos, missing values, and weird outliers, but your manager needs a baseline model today.
   • Solution: ml.clean(df) -> ml.baseline(X, y) -> Done in 5 minutes.
2. "The Ethical Challenge": You need to prove to a compliance team that your model isn't biased against a specific city or gender.
   • Solution: ml.audit(model, X, y, sensitive_features=['city']) -> Instant Model Card.
3. "The NLP Shortcut": You have 10,000 customer reviews and no time to train a classifier.
   • Solution: ml.sentiment(reviews) -> Instant insights.
4. "The Production Emergency": You have a winning model and need to hand over an API to the engineering team.
   • Solution: ml.deploy(model) -> Production code generated instantly.

---

🧠 Advanced: The "Unbreakable" Engine

What makes mlpilot different? We don't just "wrap" libraries; we harden them at the core.

AST-Healing (Self-Healing AI)

In v1.1.5, we introduced the Abstract Syntax Tree (AST) Scanner. When the AI Analyst generates code, it often includes conversational noise or syntax errors. mlpilot scans every line against Python standards and automatically purges hallucinations before they cause a crash.

Hallucination Immunity (v1.1.7)

AI models often hallucinate "setup" steps, like trying to load a file named dataset.csv. mlpilot intercepts these errors and surgically redirects the execution to the high-performance dataframe already in your memory.

Categorical Immunity (v1.2.0)

Most models crash when they see "NYC" instead of [1.0, 0.0]. mlpilot v1.2.0 introduces Categorical Guard. Functions like baseline() and tune() now automatically detect, clean, and encode string data on the fly. It just works.

The Encoding Shield (v1.2.1)

Windows legacy terminals (CP1252) often crash when a library prints an emoji (like 🚀). mlpilot v1.2.1 features a Global Encoding Shield that automatically sanitizes all output for your specific terminal in real-time, ensuring zero crashes.

Atomic Silence

We have injected a Global Silence Guard into the core initialization. No more progress-bar spam from HuggingFace, no more technical warnings from Transformers. You get only the results you asked for.

---

📦 Installation & Environments

Standard Build:

pip install mlplt

Full AI Build (Recommended):

pip install mlplt[full]

Google Colab Setup:

import mlpilot as ml
ml.colab_setup() # Automatically configures local AI backends for 100% reliability

---

🤝 Contributing & Community

Join us in building the world's most robust ML toolkit.

• Bug Tracker: GitHub Issues
• Documentation: Read the Docs

MIT © mlpilot contributors