merlin-sdk

Merlin SDK

This project contains python library for interacting with Merlin, machine learning model serving component of CaraML. It provides interface to:

• Manage ML project resources
• Track metrics and training performance
• Package and Deploy Model

Installation

pip install merlin-sdk

and authenticate to gcloud

gcloud auth application-default login

Getting Started

The easiest way to get started is to explore example notebook:

1. Deploying xgboost model
2. Deploying sklearn model
3. Deploying tensorflow model
4. Deploying pytorch model
5. Deploying user defined model

Development

Requirements:

• pipenv (pip install pipenv)
• swagger code gen (brew install swagger-codegen)

Setup development environment:

make setup

Merlin CLI

With the CLI, users can deploy a trained model (serialized model) without touching Python code at all.

To use the CLI:

// to see available subcommands
merlin --help

// to see subcommand options
merlin SUBCOMMAND --help

// deploy existing xgboost model located inside cur_model_dir
merlin deploy --url localhost --model-type xgboost --project sample --model-dir cur_model_dir --model-name xgboost-sample --env staging

// undeploy xgboost-sample version 1
merlin undeploy --url localhost --project sample --model-name xgboost-sample --model-version=1

// generate pyfunc code base
merlin scaffold -p merlin-project -m model-name -e id