End-to-end Data Science project

This is the repo with the notebooks, code, and additional material used in the ITI's workshop. The goal of the sessions was to illustrate the end-to-end process of an real project.

Additional material

In addition to the notebooks and code, the following material is also available:

Video recordings of the sessions are uploaded to youtube
Slide decks are also added to this repo here

Problem statement

Our (fictional) client is an IT educational institute. They have reached out to us has reach out with the following: “IT jobs and technologies keep evolving quickly. This makes our field to be one of the most interesting out there. But on the other hand, such fast development confuses our students. They do not know which skills they need to learn for which job. “Do I need to learn C++ to be a Data Scientist?” “Do DevOps and System admins use the same technologies?” “I really like JavaScript; can I use it in Data Analytics?” Those are some of the questions that our students ask. Could you please develop a data-driven solution for our students to answer such questions? They mostly want to understand the relationships between the jobs and the technologies.

Level guide

	Basic	Intermediate	Advanced
Business case		Decide on the KPIs that you will positively influence	Calculate the expected financial returns
Data collection	Decide on and collect a suitable data source for your business case	Decide on, collect and connect multiple data sources for better performance
Legal review		Get basic information about the local data privacy law	Study the local data privacy law
Cookie Cutter	Create the standard directory structure
Git	Use Git's GUI to track on master branch	Use Git's CLI to track on Dev branch and merge back to Master	Decide on a branching strategy and solve merge conflicts
Environments	Install python packages using conda	Create a dedicated conda environment	Share your environment and install it on a different machine
Data cleaning	Use basic statistics to filter out non-sense entries	Use advanced statistics and unsupervised learning to filter out non-sense entries	Calculate a 'sanity probability value' for each data point and use it later as the weight
Descriptive analytics	Calculate summary statistics to provide data insights	Produce visualizations to provide deeper understanding	Apply unsupervised learning to provide even deeper understanding
Predictive analytics	Create a single baseline model	Create multiple hyper-tuned models. Benchmark their performance	Combine the chosen models via ensemble and provide prediction confidence
Prescriptive analytics			Recommend the action that the user should take
Software Engineering	Refactor your notebooks to simple python scripts	Create a production OOP class for predictions	Expose your model using an API
MLops	Export and load models from pickle files	Track your models using Mlflow	Create and run a docker image for your project
Product	Create a Web App / GUI to expose prediction functionality	Add the relevant historical insights, predictions and optimization results	Collect users' feedback and retrain your model accordingly

Tutorial repo for an end-to-end Data Science project

Related tags

Overview

End-to-end Data Science project

Additional material

Problem statement

Level guide

Owner

Deena Gergis

A boosting-based Multiple Instance Learning (MIL) package that includes MIL-Boost and MCIL-Boost

Global-Local Context Network for Person Search

Code for the paper "Training GANs with Stronger Augmentations via Contrastive Discriminator" (ICLR 2021)

Equipped customers with insights about their EVs Hourly energy consumption and helped predict future charging behavior using LSTM model

Rethinking Semantic Segmentation from a Sequence-to-Sequence Perspective with Transformers

This repository contains the data and code for the paper "Diverse Text Generation via Variational Encoder-Decoder Models with Gaussian Process Priors" ([email protected])

Pyramid R-CNN: Towards Better Performance and Adaptability for 3D Object Detection

Implementation of "Learning to Match Features with Seeded Graph Matching Network" ICCV2021

This is the official pytorch implementation of the BoxEL for the description logic EL++

Adaout is a practical and flexible regularization method with high generalization and interpretability

PyTorch implementation of DeepDream algorithm

Code release for "COTR: Correspondence Transformer for Matching Across Images"

Official PyTorch implementation of Retrieve in Style: Unsupervised Facial Feature Transfer and Retrieval.

thundernet ncnn

Stochastic Downsampling for Cost-Adjustable Inference and Improved Regularization in Convolutional Networks

Implementation of paper "Towards a Unified View of Parameter-Efficient Transfer Learning"

CARL provides highly configurable contextual extensions to several well-known RL environments.

Large dataset storage format for Pytorch

MediaPipeのPythonパッケージのサンプルです。2020/12/11時点でPython実装のある4機能(Hands、Pose、Face Mesh、Holistic)について用意しています。

Course content and resources for the AIAIART course.