Ubaar Competition

Example Data Science flow for Ubaar competition

Data

The dataset comes from Kaggle Ubaar Competition

You should download it to the directory data/raw/ so that the directories structure is data/raw/ubaar-competition/*.csv

Installation

pip install -r requirements.txt

Tested using Python 3.7

Running

Process

The data science process is performed in the following steps:

• notebooks/initial_eda.ipynb - variables analysis
• notebooks/initial_training.ipynb - features impact on modeling and possible outcome
• feature_extraction/features_extraction.py - versioned scalable features extraction script
• training.ipynb - modeling in order to achieve best results

Modeling results are stored at link

Additional results

Additionally, there are 3 scripts in helper_scripts/ directory. They visualize a few aspects in transports localisations:

• helper_scripts/compare_train_test_localisations.py - assure all localisations are covered between the train and test sets (generalization on new transport sources)
• helper_scripts/prices_vis.py - average price per city visualized on a map
• helper_scripts/test_clustering.py - script for clustering analysis

Results

Results are presented in presentation Ubaar conclusions.pdf. Additionally to the model performance report, a few conclusions are drawn.

Server API

Server can be run with api/app.py

The example of running server is at http://demo-dawid.rnd.grapeup.com/ui/

and you can test it for example with a command:

curl -X POST "http://demo-dawid.rnd.grapeup.com/predict" -F "row=960218,36.666045,48.489706,زنجان,29.600574,52.537114,فارس,1092.0,751.0,treili,kafi,20.00,0"