--- title: Algoline emoji: "\U0001F537" colorFrom: indigo colorTo: purple sdk: docker app_port: 7860 pinned: false ---

## Table of Contents | Section | Description | |:--|:--| | [Overview](#overview) | What Algoline does and how it works | | [Getting Started](#getting-started) | Local setup and Docker instructions | | [Platform Walkthrough](#platform-walkthrough) | Data ingestion, exploration, training, tuning, export | | [Supported Algorithms](#supported-algorithms) | Classification and regression models | | [Architecture](#architecture) | System design, tech stack, project structure | | [Continuous Integration](#continuous-integration) | CI/CD pipeline details | | [API Reference](#api-reference) | All 31 endpoints documented | | [Contributing](#contributing) | How to report issues and contribute |
## Overview Algoline is a web-based machine learning platform that takes a raw dataset and turns it into a deployable model pipeline, entirely in the browser. There is no notebook to configure, no boilerplate to write, and no environment to set up. You upload a file, explore the data through interactive visualizations, train and compare models with a single click, optionally tune with Bayesian optimization, and download a production-ready `.pkl` file. Under the hood, a FastAPI server handles all computation. PyCaret orchestrates model comparison across scikit-learn, XGBoost, LightGBM, and CatBoost. Optuna drives hyperparameter search. Plotly renders every chart. The frontend is vanilla HTML, CSS, and JavaScript with zero build dependencies. The whole thing ships as a Docker container.
## Getting Started ### Local Setup ```bash git clone https://github.com/Al1Abdullah/Algoline.git cd Algoline pip install -r requirements.txt python main.py ``` The server starts at `http://localhost:7860`. ### Docker ```bash docker build -t algoline . docker run -p 7860:7860 algoline ```
## Platform Walkthrough ### Data Ingestion and Profiling Drop a CSV, TSV, or Excel file into the upload zone. Algoline parses it on the spot and returns four headline metrics (rows, columns, missing values, duplicates), a full statistical breakdown, inferred column types, and automated quality insights. These insights flag class imbalance, high-cardinality categoricals, constant columns, correlated features, and missing value patterns before you even ask. Select a target column and the system infers whether you are working on classification or regression. ### Exploratory Analysis Eighteen interactive chart types are available, organized into five analytical categories. Every chart is rendered with Plotly, so you can zoom, pan, hover for values, and export. | Category | Visualizations | |:--|:--| | **Distribution** | Histogram, KDE, Box plot, Violin | | **Missing Data** | Missing value bar chart, Missing value heatmap | | **Correlation** | Correlation heatmap, Pair plot, Scatter, Joint plot | | **Target Analysis** | Count plot, Pie chart, Class distribution, Target histogram, Mean target per category | | **Advanced** | Scatter vs index, Grouped box plot, Faceted small multiples | ### Model Training and Comparison Eight preprocessing toggles can be configured independently before training: | Option | Default | Purpose | |:--|:--|:--| | Drop Duplicates | On | Remove exact row copies | | Remove Outliers | Off | Filter statistical outliers | | Normalize | On | Scale numeric features | | Drop Multicollinear | On | Remove highly correlated pairs | | Transform Skew | Off | Power transforms on skewed distributions | | Feature Selection | Off | Automatic dimensionality reduction | | Polynomial Features | Off | Generate interaction terms | | Fix Class Imbalance | Off | Balance underrepresented classes | Set the test split ratio and cross-validation fold count, then train. PyCaret runs every relevant algorithm, scores each with cross-validation, and returns a ranked leaderboard. Evaluation diagnostics (confusion matrices, ROC curves, precision-recall curves, residual plots, feature importance) are generated automatically from real model predictions. ### Hyperparameter Tuning Three tuning strategies are available after initial training: | Strategy | Engine | Best For | |:--|:--|:--| | **Bayesian** | Optuna TPE | Smart, sample-efficient search | | **Random Search** | scikit-learn | Broad exploration | | **Grid Search** | scikit-learn | Exhaustive evaluation | If the tuned model does not outperform the original, Algoline preserves the original automatically. ### Export | Artifact | Format | Description | |:--|:--|:--| | **Pipeline** | `.pkl` | Serialized model ready for `predict()` in any Python environment | | **Leaderboard** | `.csv` | Full model comparison with cross-validated metrics | | **Predictions** | `.csv` | Model output on the holdout test set | Load the pipeline with `joblib` or PyCaret's `load_model()`, pass new data, and get predictions. No retraining required.
## Supported Algorithms

**Classification** Logistic Regression, K-Nearest Neighbors, Naive Bayes, Decision Tree, Random Forest, Extra Trees, Gradient Boosting, AdaBoost, XGBoost, LightGBM, CatBoost, SVM (Linear and RBF), Ridge Classifier, LDA, QDA

**Regression** Linear Regression, Lasso, Ridge, Elastic Net, Decision Tree, Random Forest, Extra Trees, Gradient Boosting, AdaBoost, XGBoost, LightGBM, CatBoost, SVR, KNN Regressor, Huber Regressor, Passive Aggressive Regressor

## Architecture ### System Design ```mermaid flowchart TD subgraph Browser["Browser"] UI["index.html + style.css + app.js"] end subgraph Server["FastAPI Server"] direction TB subgraph Routes["Route Modules"] D["data.py\nUpload and Profiling"] E["explore.py\n18 Chart Endpoints"] B["build.py\nTrain, Compare, Tune"] X["export.py\nDownload Artifacts"] end subgraph Core["Core"] H["helpers.py"] S["state.py"] end end subgraph ML["ML Engine"] PC["PyCaret 3.3"] SK["scikit-learn"] XG["XGBoost"] LG["LightGBM"] CB["CatBoost"] OP["Optuna"] PL["Plotly"] end UI -->|"fetch()"| Routes Routes --> Core B --> PC PC --> SK & XG & LG & CB B --> OP E --> PL B --> PL style Browser fill:#1e1b4b,stroke:#6366f1,color:#e4e4e7 style Server fill:#0f0d1a,stroke:#4f46e5,color:#e4e4e7 style Routes fill:#1a1730,stroke:#818cf8,color:#c4b5fd style Core fill:#1a1730,stroke:#818cf8,color:#c4b5fd style ML fill:#0c0a14,stroke:#a78bfa,color:#e4e4e7 style UI fill:#2d2a5e,stroke:#818cf8,color:#fff style D fill:#312e81,stroke:#6366f1,color:#e0e7ff style E fill:#312e81,stroke:#6366f1,color:#e0e7ff style B fill:#312e81,stroke:#6366f1,color:#e0e7ff style X fill:#312e81,stroke:#6366f1,color:#e0e7ff style H fill:#1e1b4b,stroke:#a78bfa,color:#c4b5fd style S fill:#1e1b4b,stroke:#a78bfa,color:#c4b5fd style PC fill:#312e81,stroke:#818cf8,color:#e0e7ff style SK fill:#1e1b4b,stroke:#6366f1,color:#c4b5fd style XG fill:#1e1b4b,stroke:#6366f1,color:#c4b5fd style LG fill:#1e1b4b,stroke:#6366f1,color:#c4b5fd style CB fill:#1e1b4b,stroke:#6366f1,color:#c4b5fd style OP fill:#1e1b4b,stroke:#a78bfa,color:#c4b5fd style PL fill:#1e1b4b,stroke:#a78bfa,color:#c4b5fd ``` ### Tech Stack | Layer | Technology | |:--|:--| | **Server** | FastAPI with Uvicorn | | **Runtime** | Python 3.10 | | **ML Engine** | PyCaret 3.3 (wraps scikit-learn, XGBoost, LightGBM, CatBoost) | | **Optimization** | Optuna 3.5+ with Tree-structured Parzen Estimator | | **Visualization** | Plotly 5.24 | | **Frontend** | Vanilla HTML, CSS, JavaScript | | **CI/CD** | GitHub Actions | | **Containerization** | Docker | | **Hosting** | Hugging Face Spaces | ### Project Structure ``` algoline/ ├── .github/ │ └── workflows/ │ └── ci.yml CI pipeline (lint, validate, build, smoke test) ├── app/ │ ├── __init__.py │ ├── state.py Shared session state and Plotly configuration │ ├── helpers.py Utility functions (serialization, inference, formatting) │ └── routes/ │ ├── __init__.py │ ├── data.py Upload, profiling, target selection │ ├── explore.py 18 interactive visualization endpoints │ ├── build.py Model training, comparison, hyperparameter tuning │ └── export.py Pipeline, leaderboard, and prediction downloads ├── static/ │ ├── index.html Single-page application structure │ ├── css/ │ │ └── style.css Design system (dark/light themes, glassmorphism) │ └── js/ │ └── app.js Frontend logic (navigation, API calls, chart rendering) ├── main.py Application entrypoint ├── requirements.txt Python dependencies ├── Dockerfile Production container └── LICENSE MIT ``` ### Design The interface ships with a dual-theme system. Dark mode uses glassmorphism with translucent card surfaces, ambient indigo gradients, and subtle glow effects. Light mode uses an indigo-tinted palette with layered shadows and clean gridlines. Typography is set in Inter, and every Plotly chart re-renders to match the active theme. A toggle in the top-right corner switches between the two instantly.
## Continuous Integration The CI pipeline ([`.github/workflows/ci.yml`](.github/workflows/ci.yml)) runs on every push and pull request to `main`. **Stage 1: Lint and Validate** installs dependencies, confirms the server module imports cleanly, and verifies that all expected API routes are registered on the FastAPI application. **Stage 2: Docker Build** builds the production image, starts the container, waits for the server to respond, and runs a smoke test against the root endpoint. If anything fails, container logs are captured for debugging.
## API Reference

Data Endpoints

Exploration Endpoints (18 chart types)

Training and Tuning Endpoints

Export Endpoints

## Contributing Contributions are welcome. If you find a bug or want to suggest a feature, please [open an issue](https://github.com/Al1Abdullah/Algoline/issues). For code contributions, fork the repository, create a feature branch, and submit a pull request. ## License Open source under the [MIT License](LICENSE).