Pokemon Type Predictor

Authors

Gym Leaders (Original Team)

• Caroline Tang
• Sarah Abdelazim
• Vincent Ho
• Wilfred Hass

Gym Trainers (Contributors)

Data analysis project created in part of requirements for DSCI 522 (Data Science Workflows); a course in the Master of Data Science program at the University of British Columbia. All members of this project abided by a code of conduct.

About

In this project, we attempt to classify a Pokemon's type (of which there are 18 possible types) based on the other stats (such as attack, defense, etc.) that it has. We chose k-Nearest Neighbours (k-NN) and Support Vector Classifier (SVC) algorithms for our models since they naturally support multi-class classifications, without having to use 'One-vs-One' or 'One-vs-Rest' methods. We use accuracy as the metric to score our models since there is no detriment to false positives or negatives, but we do want to know how many of the unknown Pokemon will be predicted correctly. On the unseen test data, the k -NN model predicted 60% of the new Pokemon correctly while the SVC model predicted 67% correctly. Since these are not very accurate results, we recommend trying different estimators to fill up that Pokedex!

Data

The data is found here. The data was cleaned by HansAnonymous and originally developed by simsketch. The original data can be found in the Pokemon database. All rights belong to their respective owners.

Each row in the dataset contains a different Pokemon with various attributes. The attributes are measurements of the base Pokemon, such as attack, speed or defense.The different types of Pokemon are closely related to the other attributes it possesses. For example, a rock type Pokemon is likely to have higher defensive statistics (such as defense or health points) as well as rock-type abilities. It is also most likely to be coloured grey. A complete description of each feature in the raw dataset is listed below:

Feature	Description
NUMBER	The National Pokedex Number of the Pokemon, unique to each Pokemon species
CODE	Identifier for the form of the pokemon, unique within each Pokemon species
SERIAL	Concatenation of NUMBER and CODE, unique to each row
NAME	Name of the Pokemon
TYPE1	Primary elemental type
TYPE2	Secondary elemental type, if any
COLOR	Main body color
ABILITY1	First passive ability option
ABILITY2	Second passive ability option, if any
ABILITY HIDDEN	Hidden (rare) passive ability, if any
GENERATION	Generation of games where pokemon was first introduced
LEGENDARY	Binary, indicates whether a Pokemon is legendary
MEGA_EVOLUTION	Binary, indicates whether a row is for the Mega Evolution of a Pokemon
HEIGHT	Height (m)
WEIGHT	Weight (kg)
HP	Base health point (HP) stat
ATK	Base physical attack stat
DEF	Base physical defense stat
SP_ATK	Base special attack stat
SP_DEF	Base special defense stat
SPD	Base speed stat
TOTAL	Sum of base stats

Pipeline

Report

The final report is available here.

Webpage version: https://htmlpreview.github.io/?https://github.com/UBC-MDS/pokemon-type-predictor/blob/main/doc/final_report.html

Usage

1. Without using Docker

To replicate the analysis, first clone this GitHub repository. Then, install nb_conda_kernels in your base environment. Now, install the dependencies listed in the env-poke-type-pred.yaml file below as an Anaconda environment, using:

conda install -c conda-forge nb_conda_kernels
conda env create -f env-poke-type-pred.yaml

You can switch to this environment using:

conda activate poketype

You will also need to install the R version and R packages listed here.

Then run this command:

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 details of the above commands can be found in the Makefile. This method should take at least a 1 minute to completely run.

2. Using Docker

To replicate the analysis using Docker, first clone this GitHub repository. You will also need Docker installed on your computer and have it turned on. You can follow the instructions here if you need to install Docker. After installation, navigate to the parent directory of the repository.

Note the following commands will differ depending on your system.

To reset the repository 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:

Mac (Intel)/Linux:

docker run --rm -v /$(pwd):/home/jovyan/ wthass/pokemon-type-predictor:latest make -C /home/jovyan clean

Mac (M1/M2):

docker run --rm --platform linux/amd64 -v /$(pwd):/home/jovyan/ wthass/pokemon-type-predictor:latest make -C /home/jovyan clean

Windows:

docker run --rm -v /$(pwd):/home/jovyan/ wthass/pokemon-type-predictor:latest make -C //home//jovyan clean

Then, you can reproduce the analysis using the following command:

Mac (Intel)/Linux:

docker run --rm -v /$(pwd):/home/jovyan/ wthass/pokemon-type-predictor:latest make -C /home/jovyan all

Mac (M1/M2):

docker run --rm --platform linux/amd64 -v /$(pwd):/home/jovyan/ wthass/pokemon-type-predictor:latest make -C /home/jovyan all

Windows:

docker run --rm -v /$(pwd):/home/jovyan/ wthass/pokemon-type-predictor:latest make -C //home//jovyan all

If using this method for the first time, it should take at least 8 minutes to download the Docker image and run the analysis. Any subsequent runs should take a maximum of 2 minutes.

Dependencies

• Conda Packages:
  • ipykernel
  • matplotlib
  • scikit-learn>=1.1.3
  • requests>=2.24.0
  • graphviz
  • python-graphviz
  • altair
  • altair_saver
  • selenium<4.2.0
  • pandas<1.5
  • imbalanced-learn
• Pip Packages:
  • joblib==1.1.0
  • psutil>=5.7.2
  • docopt-ng
  • vl-convert-python
• R Packages:
  • knitr
  • rmarkdown

License

The Pokemon Type Predictor materials here are licensed under the Creative Commons Attribution 2.5 Canada License (CC BY 2.5 CA) and can be found here.

Attributions

We attribute the creation of the license file to Tiffany Timbers, with more information available in the license file.

The data is attributed to the GitHub users: HansAnonymous, simsketch and the online Pokemon database.