mini-kaggle

README

Mini-Kaggle web services

A Mini-Kaggle is Kaggle inspired web service with minimalistic interface. It can be used for DataScience classes, small projects, or just for fun, to host user scores based on ML predictions.

Scores are kept in SQlite database.

Build, run instructions

To build and run the service you need Go-language and follow these steps:

# download source code
git clone https://github.com/vkuznet/mini-kaggle
cd mini-kaggle

# download appropriate SQL DB driver, e.g. SQLite
go get github.com/mattn/go-sqlite3

# download gonum package
go get github.com/gonum/stat
go get github.com/gonum/integrate
go get github.com/bsm/mlmetrics

# build the server
go build

# adjust server.json configuration file with your settings

# run the service
./mini-kaggle -config server.json

# navigate to your service, e.g. (if it runs on port 8888, see server.json)
http://localhost:8888

The configuration file serve.json should have the following structure:

{
    "uri":"sqlite3:///tmp/scores.db",
    "port": 8888,
    "templates": "templates",
    "jscripts": "js",
    "styles": "css",
    "images": "images",
    "scoreFile": "scores.csv",
    "scoreSplit": 0.7,
    "destination": "/tmp/",
    "privateKey": "123",
    "metric": "auc",
    "verbose": 0
}

Here uri defines location of SQLite scores DB, tempaltes, jscripts, styles, images define location of appropriate areas for web server. The ScoreFile points to location of score file, while ScoreSplit defines a split to divide scores into public and private sets, in this example we use 70% (0.7) for public scores and 30% for private ones. The privateKey string defines access key to private dashboard. Finally, you should specify one of the supported metrics, see metrics.go file.

Interface

Home interface provides ability to upload your score file to the service:

The Dashboard page provides basic dashboard with existing scores: