Create a ML project exposed with API

We are going to modify the data for the training format. It will be stored as a json file.

Step 0 — create basic FastApi server

In main app directory we found main.py file

Isside app folder I’ve created 3 folders

api/api_v1 —all api settings will be stored in version 1.
core —inside I will store the global configuration file
models — here I will put the files and scripts related to the training and use of the models

So, in the api/api_v1 directory I have created the api.py file which is the main point that manages all the path files.

And finally inside I have created the endpoints folder and inside I have created two (for now) files

root.py — with default response {“hello”: “world”}

predicthouseprice.py — this file will manage what to do when someone ask for house prediction.

Now we can run our server to check if the default route is working.

Step 0.1 Create Docker with FastAPI server

At this point — it’s optional, but I really like to make the configuration to use each solution as a docker image — so, let’s create a simple Dockerfile

I create Dockerfile in the main folder.

Now I can build it.

And try it.

Step 1 — Create some model

As a first model I got Predicting House Prices Challenge from HackerRank. Day 6: Multiple Linear Regression: Predicting House Prices | HackerRank

We are going to modify the data for the training format. It will be stored as a json file.

The program that fits based on this data is a simple implementation of the linear sklearn model.

We write the state of fit using pickle to the file. This file is our model. This principle is used in virtually all AI implementations.

Step 2 — Create API to query the model

First, the class we need to return prediction, based on previously trained model is like (fastapi_ml/app/models/MLR_PredictingHousePrices/predict.py)