Zero-Touch Ingestion (CSV, JSON, XML, Excel) • Regex Auto-Sanitization • Autonomous EDA • Smart NaN Imputation • 128-d Dense Tensor Synthesis • Comparative ML Model Benchmarking
The SIH 2026 AutoML Engine is an enterprise-grade, zero-touch machine learning data preprocessing and feature vectorization system built on pure Object-Oriented Programming (OOP) principles. It ingests raw, messy, and unstructured data across heterogeneous formats (CSV, Excel, JSON, XML), extracts hidden numeric values from unstructured unit strings ("1,450 sq.ft." -> 1450.0, "85 Lakhs" -> 8500000.0, "1.5 Crores" -> 15000000.0), performs autonomous EDA profiling, imputes missing values using KNN & multivariate algorithms, synthesizes a 128-dimensional dense tensor matrix (
Every major subfolder in this repository contains its own dedicated README.md detailing its underlying architecture and classes:
SIH_AutoML_Engine/
├── 📄 README.md # Main System README (This File)
├── 📄 demo_runner.py # Core Engine CLI Runner Script
├── 📁 app/ # Streamlit Web Application Dashboard
│ └── 📄 README.md # Documentation for Streamlit App Layer
├── 📁 automl_engine/ # Core OOP AutoML Package
│ ├── 📄 README.md # Core Engine Overview
│ ├── 📁 ingestion/ # Multi-format Ingestors (CSV, Excel, JSON, XML)
│ │ └── 📄 README.md # Data Ingestion Layer Documentation
│ ├── 📁 sanitization/ # Regex Auto-Sanitizer Module
│ │ └── 📄 README.md # Token Extraction & Unit Multiplier Guide
│ ├── 📁 eda/ # Autonomous EDA Profiler Module
│ │ └── 📄 README.md # Statistical Profiling Documentation
│ ├── 📁 imputation/ # NaN Imputation Layer
│ │ └── 📄 README.md # KNN & Categorical Imputation Guide
│ ├── 📁 vectorization/ # Dense Vector Synthesis Module
│ │ └── 📄 README.md # 128-d Dense Tensor Matrix (V_dense) Guide
│ ├── 📁 evaluation/ # Model Evaluator Module
│ │ └── 📄 README.md # ML Model Benchmarking Documentation
│ ├── 📁 pipeline/ # AutoML Pipeline Orchestrator
│ │ └── 📄 README.md # End-to-End Orchestrator Documentation
│ └── 📁 utils/ # Artifact Exporters & Pickle Handlers
│ └── 📄 README.md # Exporters Documentation
├── 📁 model_comparator/ # Dedicated Model Benchmarking Package
│ ├── 📄 README.md # Model Comparator Subfolder Documentation
│ ├── 📄 baseline_pipeline.py # Naive Baseline CSV Generator
│ ├── 📄 engine_pipeline.py # AutoML Feature CSV Generator
│ ├── 📄 comparator.py # Dual Pathway Evaluator (R2, MSE, MAE, RMSE, Acc)
│ ├── 📄 app.py # Dedicated Streamlit Visualization App
│ └── 📄 run_comparison_demo.py # Comparator CLI Demo Script
├── 📁 data/ # Built-in Sample Datasets (CSV, JSON, XML)
│ └── 📄 README.md # Sample Dataset Descriptions
├── 📁 output/ # Generated Processed CSVs & Exported Matrices
│ └── 📄 README.md # Exported Tensor Artifact Descriptions
└── 📁 tests/ # Automated Test Suite (100% Passed)
└── 📄 README.md # Test Suite Documentation & Execution
graph TD
A[Raw Input: CSV / Excel / JSON / XML] --> B[automl_engine.ingestion]
B -->|Raw DataFrame| C[automl_engine.sanitization]
C -->|Sanitized DataFrame| D[automl_engine.eda]
D -->|Column Classification| E[automl_engine.imputation]
E -->|Imputed Features| F[automl_engine.vectorization]
F -->|128-d Dense Matrix V_dense| G[automl_engine.evaluation]
G -->|Metrics & Artifacts| H[Streamlit Dashboard & Exporters]
streamlit run app/app.pystreamlit run model_comparator/app.pypython3 demo_runner.pypython3 model_comparator/run_comparison_demo.pypython3 tests/test_automl_engine.py
python3 tests/test_model_comparator.py-
Regression: Coefficient of Determination (
$R^2$ ), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE). - Classification: Accuracy, Precision, Recall, F1-Score (Weighted/Macro).
from automl_engine.pipeline import AutoMLPipeline
# 1. Fit zero-touch AutoML pipeline
pipeline = AutoMLPipeline(target_dim=128, imputation_strategy="knn")
cleaned_df, v_dense, report = pipeline.fit_transform("data/housing_sample.csv")
# 2. Save fitted pipeline state to .pkl file
model_path = pipeline.save_model("automl_pipeline_model.pkl")
# 3. Load fitted pipeline state from .pkl file
loaded_pipeline = AutoMLPipeline.load_model("automl_pipeline_model.pkl")