Tandur Model Deployment

Overview

This repository contains models that have been trained and are ready to be deployed. We use TensorFlow serving to deploy our model and docker to create the container from that TensorFlow serving.

Requirements

• Docker Desktop
• Windows Subsystem Linux (WSL)

Project Installation

Pull a development image (Tensorflow Serving)

Download the TensorFlow Serving Docker image and repo

docker pull tensorflow/serving

Clone the repository

git clone https://github.com/Tandur-Team/tandur-ml-deployment.git

Run a serving image

docker run -d --name serving_base tensorflow/serving

Copy /plant-model to the container's model folder

docker cp /plant-model serving_base:/models/plant-model

Commit the container

docker commit --change "ENV MODEL_NAME plant-model" serving_base plant-model

Stop serving_base container

docker kill serving_base

Run The Application

Docker will create a container from the plant-model image and run it at localhost:8501

docker run -p 8500:8500 -p 8501:8501  -t plant-model --model_config_file=/models/plant-model/models.config

API Endpoint

POST {URL}/v1/models/{plant-name}:predict : This endpoint allow user to register to the app and save user data to MySQL

• URL : Use localhost:8501 if run the applicaation on local
• plant-name : Use the same plant name as the folder name in the plant-model folder (e.g model-Padi)

Test The Application

1. Run The Application

docker run -p 8500:8500 -p 8501:8501  -t plant-model --model_config_file=/models/plant-model/models.config

2. Open POSTMAN

3. Set The API endopoint and the method

4. Request body needed to be able to run predictions

{
    "instances": [
        [
            temperature,
            humidity,
            rainfall
        ]
    ]
}

5. Click SEND

6. You should get a JSON response