go>> ML-Model-Deployment>> 返回
项目作者： sumitkutty

项目描述 ：
Deployment of a machine learning model using Flask and Heroku.
高级语言： Python
项目地址: git://github.com/sumitkutty/ML-Model-Deployment.git
创建时间： 2021-01-31T17:31:28Z
项目社区：https://github.com/sumitkutty/ML-Model-Deployment
开源协议：
下载


ML-Model-Deployment

This project is a demonstration of a machine learning model deployment using Flask and Heroku on a simple dataset.

App Link: https://car-sp-prediction-app.herokuapp.com

Objective:

Predicting the selling price of a car given features like kms driven, transmission type, year bought, present price etc.

Dataset:

The dataset’s was collected from cardekho.com was provided bdataset.

Link: https://www.kaggle.com/nehalbirla/vehicle-dataset-from-cardekho

  • No of features: 9
  • No of observations: 301

Packages:

  • Numpy, Pandas, Matplotlib, Flask, Pickle, sklearn

Preprocessing:

  • Null Values : There are no null values in the dataset.
  • Feature Elimination: The ‘Car_Name’ feature was removed as it would not contribute in a beneficial way. The ‘Year’ feature was also removed.
  • Feature Engineering: The ‘car_age’ feature was computed by calculating the difference between ‘Year’ variable and current year.
  • Encoding: The categorical features ‘Fuel_Type’, ‘Seller_Type’, ‘Transmission’ were one-hot encoded.

Predictive Modelling:

  • ML ALgorithms: Random Forest was used for the training process.

Evaluation:

  • Metrics: R2, adjusted R2, negative mean squared error.

Performance (Random Forest):

  • R2: 0.9477
  • adjusted R2: 0.9300
  • neg mean squared error: -2.44

Productionisation:

App:

Flask: Flask is a simple to use web framework for python. It can be used to make web apps, and deploy machine learning models to production.
Steps:

  • Created an app to to load a HTML UI when a user visits a url.
  • The GUI contains input spaces for the user to enter values into.
  • The app runs another function which preprocesses and predicts the output for the inputs entered by the user.
  • The app then returns the output along with a html file to the UI.

Deployment:

Heroku: Heroku is a cloud based PaaS platform. It is used to operate and work on applications directly on the cloud. Heroku was used in this project to deploy the flask app to the web.
Steps:

  • A wsgi file is created as an interface between a application and a web server.
  • Created a process file for heroku execution.
  • Pushed the entire app to the git remote repository of heroku.
  • The url of the app was created.