Uncertainty Dashboard

Our visualization proposes a reusable representation of prediction systems to guide decision makers to understand the implicit uncertainty that is present on machine-learning prediction algorithms.

In particular, for our case of study (Academic Risk based on Historical Data). There are three main sources of uncertainty, the prediction model, the data consistency and the case completeness of the historic dataset.

Uncertainty Dashboard is a proposed visualization technique where the risk to fail in an academic semester is predicted by the machine learning algorithm and is presented to the users (professors and academics) in a straightforward way, so they can make an informed decision concerning the optimal student's academic load.

## Early Prototype

How to

Be sure to install meteor.js in your machine, then simply:

git clone https://github.com/FranciscoGutierrez/uncertaintyDashboard.git
cd uncertaintyDashboard
meteor

URLs for testing:

// anibal
// http://localhost:3000/200909893?c=ICM00166,FIEC03046,FIEC01545
//
// roger
// http://localhost:3000/200834711?c=ICM00166,FIEC03046,FIEC01545

Technical Documentation

Architecture Overview

The following are useful scripts for the development/deployment of the dashboard.

Dealing with database

# Going into the database and get some data
meteor mongo
show dbs
db.courses.find()
# Destroy database, reload project dependencies and stuff from scratch.
meteor reset

Import Data

Basic mongoimport command.

mongoimport --db meteor --collection courses --file courses.json --host=127.0.0.1:3001
mongoimport --db meteor --collection studentscourses --file students_courses.json --host=127.0.0.1:3001
mongoimport --db meteor --collection students --file students.json --host=127.0.0.1:3001

Python Converters

If data is on CSV this is an easy way to convert it to JSON, and then import.

Option 1

import csv
import json

csvfile = open('courses.csv', 'r')
jsonfile = open('courses.json', 'w')

fieldnames = ("alpha","beta","skewness","_id","name")
reader = csv.DictReader( csvfile, fieldnames)
for row in reader:
    json.dump(row, jsonfile)
    jsonfile.write('\n')

Option 2

import csvmapper
jsonfile = open('courses.json', 'w')
# parser instance
parser = csvmapper.CSVParser('courses.csv', hasHeader=True)
# conversion service
converter = csvmapper.JSONConverter(parser)
print converter.doConvert(pretty=False)
jsonfile.write(converter.doConvert(pretty=False))

Communication

Clients use WebSocket as communication channel with the prediction server, when asking for a prediction the client uses the following format:

[
  {
    "requestId": "5645f7f7ef0bde57344c84de",
    "student": [
      {
        "id": "5645f7f70a7cd6ab830e8722",
        "gpa": 7.0793,
        "performance": 0.6,
        "compliance": 3
      }
    ],
    "courses": [
      {
        "id": "5645f7f73077f65af76df45a",
        "compliance": 2
      },
      {
        "id": "5645f7f7fccab1b5dd662a4b",
        "compliance": 5
      },
      {
        "id": "5645f7f7e7b8050367ef5cff",
        "compliance": 5
      },
      {
        "id": "5645f7f70ea9b237c0039330",
        "compliance": 5
      }
    ],
    "data": [
      {
        "from": 2009,
        "to": 2015,
        "program": true,
        "sylabus": true,
        "evaluation": false,
        "instructors": true,
        "compliance": 2
      }
    ]
  }
]

To change the prediction server URL there is a special file named websocket.js at the client folder.