Kaggle Competition shelter-animal-outcome
This is my project for the shelter-animal-outcome competition on Kaggle.
My rank in the competition is 79th / 1604 (https://www.kaggle.com/competitions/shelter-animal-outcomes/leaderboard)
To run the script, the following Python modules need to be installed:
numpy / pandas / scikit-learn / xgboost
You need to place train.csv, test.csv, and submission.csv in the data/ directory, or replace the following directory paths in conf/simple.conf with the actual paths to train.csv, test.csv, and submission.csv：
train_filename=./data/train.csv
test_filename=./data/test.csv
submission_filename=./data/submission.csv
Then, run shelter.py in the bin/ directory:
python bin/shelter.py
The following configurations in conf/simple.conf control the operations:
do_train=1 # whether to train
do_validation=1 # whether to perform validation
do_search_parameter # whether to perform parameter search (needs corresponding parameter range specified in the code)
do_test=1 # whether to test
Project file structure:
bin/
simple.py / shelter.py: related to the final version of the project results
lib/
*: related to the final version of the project results
bin.bk/
*_simple.py, model_average.py: historical attempts, including individual usage of random forest, KNN, XGBoost, and model fusion attempts
For more information about the project, please refer to my blog.
I have kept all historical versions in the bin.bk/ directory without making any formatting corrections. The main reason is that I didn't have a suitable model training template in the early stages to try different model methods. After unifying various model methods later on, I developed my model trainer template method, which was used in the final version submission. 关于项目的介绍请见我的博客。