IRIS

An example of a neural network

About

This project fetches the IRIS dataset from github and trains a sequential neural network to classify the different flower species in the dataset.

Install node-red

This project depends on node-red. There is a fork of the project with a working docker container here

First thing is to clone this repository and build the docker on the host machine. You can decide a port through the envorinment variable.

$ git clone https://github.com/jolin1337/nodered-ml
$ cd nodered-ml
$ PORT=1880 docker-compose up

Once the docker is built and up and running you should be able to navigate to http://localhost:1880/admin and setup this repository in ./projects/IRIRS to get started.

Use node-red interface

To be able to experiment with the setup make sure to get familiar with how node sends messages between nodes here: https://nodered.org/docs/user-guide/messages

The fun parts

This project consists of 4 flows; Data processing, Model training, Evaluation and Deployment

• The Data processing flow makes sure the data is available in the node-red environment and prcessing the data accordingly to what the model input needs. To get an overview of the dataset you can go to http://localhost:1880/red-nodes/ui after this flow has been executed.
• The Model training flow you can tweak parameters and execute a training process that will output a model that you can use in evaluation and deployment flows later for evaluation and predictions
• The Evaluation flow you are able to do sample predictions and evaluate the model agains an test dataset. After this flow is executed an evaluation website will be available to view the results in different formats at http://localhost:1880/red-nodes/ui
• The deployment flow is enables you to offer predictions to 3rd party services. The 3rd party services can for example ask node-red service about a flower type.