DeepChem Server Documentation
Welcome to DeepChem Server, a minimal cloud infrastructure for DeepChem that provides a FastAPI-based backend for managing datasets, running featurization tasks, and building machine learning models with DeepChem.
DeepChem Server offers a streamlined way to:
Upload and manage datasets in various formats
Perform molecular featurization using DeepChem’s extensive featurizer library
Train machine learning models and run inference on molecular data
Evaluate model performance with comprehensive metrics
Store and retrieve models and data through a unified datastore API
Access functionality through both REST API endpoints and a Python client library
Quick Start
To get started with DeepChem Server:
Installation: Clone the repository and run the server using Docker
Upload Data: Use the API or Python client to upload your datasets
Use Core Primitives: Transform data, train models, and run inference using the four main primitives
Deploy Models: Access functionality through REST API endpoints
Server Setup
The fastest way to get started is using Docker:
git clone <repository-url>
cd deepchem-server
bash docker.sh
This will start the server on http://localhost:8000.
Key Features
FastAPI Backend: Modern, fast web framework with automatic API documentation
DeepChem Integration: Built-in support for molecular featurization and modeling
Flexible Storage: Disk-based datastore with support for various data formats
Python Client: Easy-to-use Python library for programmatic access
Docker Support: Containerized deployment for easy setup and scaling
Architecture Overview
DeepChem Server consists of several key components:
Core Primitives: Main functions for featurization, training, inference, and evaluation
API Layer: FastAPI routers handling HTTP requests
Core Modules: Business logic for data handling, featurization, and model management
Datastore: Abstract storage layer with concrete disk implementation
Client Library: Python SDK for easy integration