An open API service indexing awesome lists of open source software.

https://github.com/asherk7/auto-ml

Creating a basic Auto-ML system
https://github.com/asherk7/auto-ml

ml swe

Last synced: 8 months ago
JSON representation

Creating a basic Auto-ML system

Awesome Lists containing this project

README

          

# AutoML System

A lightweight, modular AutoML framework for computer vision and tabular machine learning. Supports training, experiment tracking, and model serving with minimal setup.

---

## Features

* **Tasks**:

* Computer Vision → image classification, object detection
* Tabular → classification, regression, clustering

* **Models**:

* CV → ResNet, EfficientNet, MobileNet (PyTorch Lightning)
* Tabular → Random Forest, Logistic Regression, SVM, KNN, Decision Trees (scikit-learn)

* **Core**:

* Auto preprocessing + augmentation
* Hyperparameter tuning (GridSearchCV)
* Experiment tracking (Weights & Biases)
* ONNX export for inference
* FastAPI serving + SQLite metadata DB
* Docker ready

* **Datasets**: CIFAR-10, Iris, + custom datasets

---

## Quick Start

```bash
# Clone & install
git clone https://github.com/asherk7/auto-ml.git && cd auto-ml
pip install -r requirements.txt

# Check system info
python main.py system-info
```

### Examples

```bash
# Train CV model (ResNet-18 on CIFAR-10)
python main.py train-cv --model-type resnet18 --max-epochs 10

# Train tabular model (Random Forest on Iris)
python main.py train-tabular --model-type random_forest --tune-hyperparameters

# Try multiple models automatically
python main.py auto-train --data-type tabular --task-type classification --models "random_forest,svm,logistic_regression"

# Start API server
python main.py serve --port 8000
```

---

## Project Structure

```
auto-ml/
├── main.py # CLI entry
├── models/ # CV + tabular models
├── training/ # Training logic
├── serving/ # FastAPI endpoints
├── utils/ # DB + helpers
└── examples/ # Example scripts
```

---

## Key Commands

| Command | Use |
| ------------------ | -------------------- |
| `train-cv` | Train vision model |
| `train-tabular` | Train tabular model |
| `auto-train` | Run multiple models |
| `serve` | Start FastAPI server |
| `list-experiments` | Show experiments |
| `list-models` | Show trained models |

---

## Advanced

* W\&B for experiment tracking (`wandb login`)
* Export to ONNX for fast inference
* Docker & docker-compose for deployment

---

## Roadmap

* Optuna for smarter hyperparameter tuning
* Fine-tuning large pretrained models (QLoRA, PEFT)
* Explainability → GradCAM, SHAP/LIME
* Redis + Celery for async serving
* Interactive model builder
* Real-time deployment for latency and inference testing