README.md

karura

karura enables you to use machine learning automatically & interactively.

Architecture

karura has insights.

Each insight gets the data and judges the necessity of its adoption, and if it needed, execute it.

For example, NAFrequencyCheckInsight watches the amount of the NA in each column, and if it is too high, then drop the column. Of course, you can confirm it to the user.

karura can have many insights, so you can add the insight as you needed.

Insights are adopted according to the InsightIndex order.
And you can create custom insight by inheriting the Insight class.

karura is multi-language application. Now supports ja and en.
(Some message on kintone is only Japanese).

Usage

In the Jupyter Notebook

You can use karura as your partner for data analytics.

karura notebook

To install karura, pip install.

pip install karura

The dependencies as followings.

• numpy
• scipy
• scikit-learn
• matplotlib
• pandas

If you use Slack integration, additionally install below.

• slackbot

If you use kintone integration, additionally install below.

• pykintone
• tornado
• cryptography
• pymongo (Also needs MongoDB)

As Slackbot

You can communicate with karura on Slack!

When you upload the csv file or tell kintone app name to karura, then interaction starts.You can build your own machine learning model interactively, and also you can get some suggestions about the data treatment from karura.

As Adviser on kintone

You can ask karura to analyze your kintone app!

• Select the target app
• Select the field that you want to predict and fields that you use to do it
• Push Train button

Then, you can get analyzed result!

Setup

Slackbot

• Use Dockerfile_slackbot
• set below environmental variables
  • SLACK_TOKEN: Your Slack token
  • LANG: language that you want to use (ja or en)

kintone

Tutorial is available (ja)