search

UBC-MDS / Coffee_quality_rating_predictor

Coffee Rating Prediction for DSCI 522
MIT License
1 stars 8 forks source link

readme

Coffee Quality Predictor

Creators: Kristin Bunyan, Arlin Cherian, Berkay Bulut, Michelle Wang

Introduction

This data analysis was performed as part of the DSCI 522, Data Science Workflows course at the University of British Columbia's Master of Data Science Program. Our team will use machine learning techniques to predict coffee quality on arabica coffees. We will be using an exploratory data analysis roadmap and predictive modeling to create this project.

About

Summary

In this analysis, we attempt to find a supervised machine learning model which uses the features of the Coffee Quality Dataset, collected by the Coffee Quality Institute in January 2018, to predict the quality of a cup of arabica coffee to answer the research question: given a set characteristics, what is the quality of a cup of arabica coffee?

We begin our analysis by exploring the natural inferential sub-question of which features correlate strongly with coffee quality, which will help to inform our secondary inferential sub-question: which features are most influential in determining coffee quality? We then begin to build our models for testing.

After initially exploring regression based models, Ridge Regression and Random Forest Regressor, our analysis deviated to re-processing our data and exploring classification models. As you will see in our analysis below, predicting a continuous target variable proved quite difficult with many nonlinear features, and was not very interpretable in a real sense of what we were trying to predict. Broadening the target variable and transforming it into classes: “Good” and “Poor”, based on a threshold at the median, helped with these issues.

Our final model, using Random Forest Classification, performed averagely on an unseen test data set, with an ROC score of 0.67. We recommend continuing to study to improve this prediction model before it is put to any use, as incorrectly classifying the quality of coffee could have a large economic impact on a producers income. We have described how one might do that at the end of our analysis.

Dataset

We will be analyzing the Coffee Quality Dataset, collected by the Coffee Quality Institute in January 2018. The data was retrieved from tidytuesday, courtesy of James LeDoux, a Data Scientist at Buzzfeed (DeLoux, J). The data is collected on Arabica coffee beans from across the world and professionally rated on a 0-100 scale based on factors like acidity, sweetness, fragrance, balance, etc. The dataset also contains information about coffee bean origin country, harvesting and grading date, colour, defects, processing and packaging details. For a full description of the variables included in the dataset, please visit the website linked above.

Report

The final report can be found here.

Usage

There are three ways to run the analysis.

1. Using shell script

To replicate the analysis, do the following:

  1. clone this GitHub repository
  2. create a conda environment with all the dependencies using the environment.yaml file with your terminal: 
    conda env create -f environment.yaml
    conda activate 522_group_03
  3. run this script to install the R-dependencies: 
    Rscript -e 'install.packages("knitr", repos="https://cloud.r-project.org")'
Rscript -e 'install.packages("kableExtra", repos="https://cloud.r-project.org")'
Rscript -e 'install.packages("rmarkdown", repos="https://cloud.r-project.org")'
  4. Run the following commands at the command line/terminal from the root directory of this project:
# download data
python src/download_data.py --url=https://raw.githubusercontent.com/rfordatascience/tidytuesday/master/data/2020/2020-07-07/coffee_ratings.csv --out_file=data/raw/coffee_ratings.csv

# pre-process data
python src/prepare_data.py --input_data=data/raw/coffee_ratings.csv --out_dir=data/processed/

# run eda analysis
python src/plot_visualisations.py --input_data='data/processed/train_df.csv' --out_dir='results/images/'

# run the main analysis
python src/ml_analysis.py --train="data/processed/train_df.csv" --test="data/processed/test_df.csv" --table_file="results/model_comparison.csv" --out_dir="results/images/"

# render final report
Rscript -e "rmarkdown::render('reports/coffee_rating_prediction_report.rmd', output_format = 'html_document')"

The estimated time to download data and perform analysis via shell script is less than 5 min.

2. Using Make

To replicate the entire analysis and output the final report, clone this GitHub repository, install the dependencies listed below (or create a conda environment with all the dependencies using the environment.yaml file as described above) and run the following command at the command line/terminal from the root directory of this project:

make all

To reset the repo to a clean state, with no intermediate or results files, run the following command at the command line/terminal from the root directory of this project:

make clean

The estimated time to download data and perform analysis via makefile is less than 5 min.

3. Using Docker

note - the instructions in this section also depends on running this in a unix shell (e.g., terminal or Git Bash)

To replicate the analysis, install Docker. Then clone this GitHub repository and run the following command at the command line/terminal from the root directory of this project:

docker run --rm -v /$(pwd)://home//rstudio//coffee berkaybulut/coffee_prediction:v0.7.0 make -C //home//rstudio//coffee all

To reset the repo to a clean state, with no intermediate or results files, run the following command at the command line/terminal from the root directory of this project:

docker run --rm -v /$(pwd)://home//rstudio//coffee berkaybulut/coffee_prediction:v0.7.0 make -C //home//rstudio//coffee clean

The estimated time to download data and perform analysis via docker is around 5 min.

Dependency Diagram of Makefile

Dependency Diagram

Dependencies

References

DeLoux, J. "coffee-quality-database" June 2018. https://github.com/jldbc/coffee-quality-database

Coffee Quality Institute's review pages, January 2018. https://database.coffeeinstitute.org/