Good Grades Predictor

Evolution and Learning in Computational and Robotic Agents
MSE 2400   Dr. Tom Way

Description

• This activity enables you to use a neural network to select classes to take where you are likely to get a good grade.

Getting Started

• If you are running the neural network software for this exercise, install JustNN if you haven't already

• Download the starter net GoodGrades.tvq
• Open GoodGrades.tvq in JustNN
• Notice that the example has a number of classes that Dr. Way took, including how he did in each course.

Training

• Add more course information of your own as training or validation data, using the existing columns:
  • First column - name of person taking class
  • Subject - brief name of class
  • Lk teacher? - did you like the teacher? true or false
  • Lk subject? - did you like the subject? true or false
  • Friend? - did you have a friend in the class with you? true or false
  • Time - time of day class met (1 = morning, 2 = midday, 3 = afternoon, 4 = evening
  • Hard - how hard was the class on a scale of 0=super easy, 10=extremely hard
  • Worked - how hard did you work in the class, 0=not at all, 10=extremely hard
  • Is boring? - was the class boring? true or false
  • Is fun? - was the class fun? true or false
  • Learn lots? - did you learn a lot? true or false
  • Days/week - how many days per week did the class meet?
  • Missed % - what percentage of class meetings did you miss or skip?
  • Grade (%) - what point grade did you earn (0-100)
  • Good Grade? - did you get a "good" grade? true or false
• Train the network

Testing

• Using the same columns of data, add some querying rows for classes you are thinking of taking
• This time, because you are creating querying rows, you won't enter values for Grade (%) or Good Grade?
• If your neural network is well-trained, the output values for your querying rows will be filled in automatically!
• Look at the output of your querying rows to see if the network thinks you should take the class or not!

Discussion

• How accurate were the predictions?
• If you redefine some classes you did take in the past as querying rows, how does it do?
• Is this a good way to select classes?
• What data is missing in predicting or selection which classes to take?