Contents

This repository contains a training pipeline for the iris-dataset classification and a REST API for model utilization that can be consumed by other services.

Instruction on how to reproduce the environment for running both the pipeline and the API:

The repo contains a file denoted 'requirements.txt'. You can create your Python virtualenv and then run the following command:

• $ pip install -r requirements.txt

ML model Training pipeline for IRIS

The pipeline is written in Python and performs the following steps:

• Missing data imputation
• Normalization
• Model fitting with GridSearchCV

How to run the training pipeline:

• The training pipeline is implemented on file modeling/modeling.py
• For running in a Python environment, run the following command:
  • $ python modeling/modeling.py

The model chosen for the application was the LogisticRgressor from sklearn, and the training process performs a cross-validation with 5 folds, searching over a domain of 3 parameters.

REST API for model utilization

For the model utilization, a REST API was implemented. The API is written in Python and uses the Flask framework. The API contains a POST endpoint for performing the prediction under the route: classify/iris

A validation stage was also implemented. Also, a set of unitary tests written with Pytest is available.

How to init the API:

• For running in a Python environment, run the following command:
  • $ python app/api.p

How to request to the API:

• The API accepts only POST requests, in the route: classify/iris for performing the Iris classification. The method receives a JSON in the following format: { "sepal_width": 1.0, "sepal_length": 1.5, "petal_length": 1.0, "petal_width": 0.5 }

The values for each of the four arguments must be float or int.

• The API returns a response JSON containing the classification as shown in the example below: { "iris-classification": "Iris-setosa" }

• An example snippet for requesting from a Python script is shown below:

  import requests data = { 'sepal_width': 1.0, 'sepal_length': 0.5, 'petal_length': 2.0, 'petal_width': 2.5 } res = requests.post(url="http://127.0.0.1:8080/classify/iris", json=data)

• It is also possible to request from a GUI client such as Insomnia, Postman, or using curl:

• e.g: curl -d '{"sepal_width":1.0, "sepal_length":0.5, "petal_length": 2.0, "petal_width": 2.5}' -H "Content-Type: application/json" -X POST http://localhost:8080/classify/iris

Extras

The repository also contains a Dockerfile for the API, that can be used to deploy the application on a server For building the image, run:

• docker build . -t api:v1 For running the image:
• docker run api:v1