ml-service

Requirements

1. Poetry osx / linux / bashonwindows install instructions:

curl -sSL https://raw.githubusercontent.com/python-poetry/poetry/master/get-poetry.py | python -

Check more details in Poetry Docs.

2. RClone

To install rclone on Linux/macOS/BSD systems, run:

curl https://rclone.org/install.sh | sudo bash

Check more details in RClone Download.

Tutorial

Install

1. Install project with poetry

poetry install

Train model

1. Run MlFlow Server

make run-mlflow

2. Train a model with given script.

poetry run python tutorial/train.py

Download model from model registry

1. Go to mlflow ui page and check its run id

http://localhost:5000

1. Download model to local with ml-service sync.

poetry run ml-service sync --run-id=...

3. Check download is complete.

ls mnt/artifacts/model/

If completed you'll get result as below

MLmodel          conda.yaml       model.pkl        python_env.yaml  requirements.txt

Run FastAPI Server

1. Run fastapi server with ml-service server.

poetry run ml-service server --model-name "model" --artifact-path "mnt/artifacts"

2. Check fastapi docs site:

http://127.0.0.1:8000/docs

After deployment

1. Go to FastAPI docs site:

2. Check engine post

3. Write predict_method to use, in this tutorial we use predict. And click excute.

4. After excute, you can get result as below:

Tutorial with buffer

Buffer domain save input data to postgres server.

Requirements

1. postgresql

Install postgresql with instruction [LINK].

Setup

1. init db folder to start:

make initdb

2. run postgresql server:

make run-postgres

3. add user to postgresql server:

make init-user

Run FastAPI Server

1. Run fastapi server with ml-service server with --use-buffer argument.

poetry run ml-service server --model-name "model" --artifact-path "mnt/artifacts" --use-buffer

After deployment

1. Use buffer domain and execute.

Check DB

1. Connect to DB with below information:

   • db: postgres
   • user: postgres
   • password: password

2. Check input table: