- Robin Dhillon
- Lisha Gao
- Markus Nam
- Eyre Hong
This data science project is created for DSCI 522 (Data Science Workflows); a course in the Master of Data Science program at the University of British Columbia.
Chocolate is a treat that is beloved globally. Different types of chocolates and their manufacturers are prevalent in different parts of the world. Hence, we decided to explore what features give us the best ratings for these various chocolates. Specifically, do features like Manfacturer, Cocoa Percent and Country of Bean Origin affect or influence the ratings of one chocolate over another? The main goal of this project is to explore this question using different machine learning models to predict chocolate ratings. Therefore, we perform an analysis based on a dataset of 2,588 chocolate ratings compiled by Brady Brelinsk from the Manhattan Chocolate Society. The dataset is available to the public here.
Since we have a predictive research question, we separate the dataset into train data and test data (75% and 25%, respectively) to avoid violation of the Golden Rule. Initial exploratory data analysis (EDA) includes investigating the distributions of the aforementioned features as well as features such as the Ingredients and Characteristics of the chocolates. Plots such as barplots and boxplots are created of these features and saved in the results folder. The corresponding dataframes are saved as csv files in the results folder as well. As we delve deeper into the analysis, other plot types will most likely be used as well.
When we create the machine learning models, we preprocess the data using StandardScaler(), OneHotEncoder(), and CountVectorizer() to account for the numeric, categorical, and text columns of our dataset, respectively. For the predictions, we use regression models such as Ridge and SVR (Support Vector Regression) while using DummyRegressor() as the baseline. These models are fit to the training data and then evaluated by cross-validation. The scoring metric used is neg_mean_absolute_percentage_error. The models are further improved via hyperparameter optimization to obtain the best model, which are then used for predictions on the test data. The hyperparameter optimization is done via RandomizedSearchCV. Ultimately, these models are saved as sav files in results. In the end, we check how our best model scores on the test data and the output is saved as csv files.
The workflow of how the data analysis should be run is shown in the flowchart/diagram below and the instructions are shown in the Usage:
Figure 1. Flowchart of analysis
Below, we suggest three different ways to run this analysis:
Note - the instructions in this section depends on running this in a unix shell (e.g., terminal or Git Bash). Specifically, these commands should be run on 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 to pull the docker image:
docker pull eyrexh/chocolate_rating_docker:v0.1.0
For Mac computers with an Intel processor, activate the docker image and run our project by using:
docker run --rm -v "/$(pwd)":/home/chocolate eyrexh/chocolate_rating_docker make -C /home/chocolate all
If results already exist, or if the results of the above command need to be undone for a clean state of the repository, run the following command at the command line while still being at root directory of this project:
docker run --rm -v "/$(pwd)":/home/chocolate eyrexh/chocolate_rating_docker make -C /home/chocolate clean
Note: If you are using M1 / M2 mac platform, please add the flag --platform linux/amd64 when running the above docker run command.
i.e.
docker run --rm --platform linux/amd64 -v "/$(pwd)":/home/chocolate eyrexh/chocolate_rating_docker make -C /home/chocolate all
AND
docker run --rm --platform linux/amd64 -v "/$(pwd)":/home/chocolate eyrexh/chocolate_rating_docker make -C /home/chocolate clean
For Windows computers, use the following command to activate the docker image and run our project:
docker run --rm -v "/$(pwd)"://home//chocolate eyrexh/chocolate_rating_docker make -C //home//chocolate all
For a clean repository, use the following command:
docker run --rm -v "/$(pwd)"://home//chocolate eyrexh/chocolate_rating_docker make -C //home//chocolate clean
If Docker is unavailable, we suggest using this method to replicate this analysis. First, please clone this repository and install the dependencies by referring to the Installation section. Next, while you are at the the root directory of this project, run the following command at the command line:
make all
If results already exist when make all is used, or if the results of the above command need to be undone for a clean state of the repository, run the following command at the command line while still being at root directory of this project:
make clean
Install the dependencies listed in dependencies, run the command shown below from the root directory of this project to download the raw data, and then run the EDA file on JupyterLab.
-
To download the data file:
python src/download_data.py --url=http://flavorsofcacao.com/database_w_REF.html --out_file=data/raw/chocolate_raw.csv -
To perform data cleaning, and split raw data file into train set and test set:
python src/data_preprocessing.py --in_file=data/raw/chocolate_raw.csv --out_dir=data/processed/ -
To generate Exploratory data analysis (EDA) output:
python src/rating_eda.py --in_file=data/processed/train.csv --out_dir=results/ -
To train baseline model:
python src/model_baseline.py --in_file=data/processed/train.csv --out_dir=results/ -
To train SVR model:
python src/model_svr.py --in_file=data/processed/train.csv --out_dir=results/ -
To train Ridge model:
python src/model_ridge.py --in_file=data/processed/train.csv --out_dir=results/ -
To generate score of the models from test data set:
python src/model_summary.py --in_file=data/processed/test.csv --model_dir=results/ --out_dir=results/ -
To generate the final analysis report:
Rscript -e "rmarkdown::render('doc/chocolate_rating.Rmd')"
The final report can be found here in the doc folder.
Create a conda environment by running the command below:
conda env create -f src/env-chocolate.yaml
To activate the environment, please run:
conda activate chocolate
Install the necessary R packages:
Rscript -e "install.packages(c('knitr', 'kableExtra', 'tidyverse', 'caret', 'xfun'), repos='https://cran.rstudio.com/')"
- ipykernel
- matplotlib>=3.2.2
- scikit-learn>=1.1.3
- requests>=2.24.0
- graphviz
- python-graphviz
- eli5
- shap
- jinja2
- altair_saver
- selenium<4.3.0
- pandas<1.5
- imbalanced-learn
- pandas-profiling
- ipywidgets
- lxml
- pandoc
- pip
- lightgbm
- pip:
- joblib==1.1.0
- mglearn
- psutil>=5.7.2
- vl-convert-python
- docopt-ng
The materials here for Chocolate Rating Predictor are licensed under the Creative Commons Attribution 2.5 Canada License (CC BY 2.5 CA). MIT License
-
Brady Brelinski and Andrea Brelinski. 2022. "chocolate_database" Flavor of Cacao, Mahanttan Chocolate Society http://flavorsofcacao.com
-
Roger D. Peng and Elizabeth Matsui. 2017. "The Art of Data Science" https://bookdown.org/rdpeng/artofdatascience/