NeuralNetworkTester

This program is designed to help you with homework 2 by testing your networks and visualizing the results. Besides using an equation, you can also define your own underlying function the network should learn by drawing it. After generating an input file this tool will test your compiled C program and compare its results to the ground truth data by plotting your predictions. You can use this script for regression (including the bonus problem) and classification tasks.

Installation

1. Install Python 3 on your machine

   • Windows: Download "Windows x86-64 executable installer" listed under the latest Python 3 version on this website: https://www.python.org/downloads/windows/ Follow the installation process and make sure that you include pip and that you add Python to the PATH variable during the installation. You can also find a video explanation on how to install Python 3 on YouTube: https://youtu.be/dX2-V2BocqQ
   • Ubuntu: Probably Python 3 is already installed, if not use the following command:

     sudo apt-get install python3 
     

     The same holds for pip3. Usually it should already be installed, otherwise write:

     sudo apt-get install python3-pip
     

   • Mac: Download "Mac OS X 64-bit/32-bit installer" listed under the latest Python 3 version on this website: https://www.python.org/downloads/mac-osx/ Follow the installation process and make sure that you include pip during the installation. By default that choice is already set. You can also find a video explanation on how to install Python 3 on YouTube: https://youtu.be/uA8SA81nivg

2. Download the file main.py and save it at a desired location

3. Run the file main.py
   If you're on Windows and want to run it from the command line, switch to the directory where you put the file and type

   python main.py
   

   On Linux or Mac, change to the right directory and write

   python3 main.py
   

4. Maybe you get an error because you're missing some modules. The following modules are required:

   • numpy
   • scipy
   • matplotlib

   Install them for example with pip. If you are on Windows, open the command line and type

   pip install moduleName
   

   if you use Linux or Mac, open the terminal and write

   pip3 install moduleName
   

   Replace moduleName with the name of the module you are missing.

   On Windows, the installation of scipy often causes problems. Instead of installing it directly via pip, follow these steps:

   1. Go to https://www.lfd.uci.edu/~gohlke/pythonlibs/#numpy and download the NumPy file that corresponds to your configuration. The two digits after "cp" in the name specify the Python version you have installed and "win32" or "amd_win64" correspond to a 32bit or a 64bit version of Python.
   2. Open the command line and change to the directory where you have downloaded the file.
   3. Run

      pip install numpy‑1.13.3+mkl‑cpxx‑cpxxm‑xx.whl
      

      where you have to use the name of the file you just downloaded. After that, you should have sucessfully installed NumPy.
   4. Go to https://www.lfd.uci.edu/~gohlke/pythonlibs/#scipy and download the SciPy file that corresponds to your configuration.
   5. Open the command line and change to the directory where you have downloaded the file.
   6. Run

      pip install scipy‑1.0.0‑cpxx‑cpxxm‑xx.whl
      

      where you have to use the name of the file you just downloaded. After that, you should have sucessfully installed SciPy.

Usage

To use this tool, simply run the file main.py. The following window opens:

In the lower part you can define the parameters of the model you want to run

• Mode: Choose whether you want to do a classification or a regression task
• Data source: Specify if an equation should be used for the underlying function or whether you want to use your own drawn function
• Noise: Only accessible if mode is set to "Regression". The training data is gaussian distributed around the true value. Here you specify the standard deviation
• Number points training: How many training samples should be generated
• Number points test: How many test samples should be generated
• Time limit (s): After your program runs this number of seconds, it will automatically terminate. Note that this is also what happens on the server used to grade your homework
• Compiled C program: Select the compiled C program for either classification or regression depending on your earlier choice.
  NOTE: do not select the C code (a *.c file) here in contrast to the grading system

Equation

If you want to use an equation for the underlying function, choose "Function" for the value of Data source.
You specify everything that is related to this function in the upper left area of the window.

• Function y(x)=: Give the equation of the function that should be used, e.g. 7+exp(-x/10)*cos(x)**2
  Use Python syntax for the equation. You can use the following functions and constants: sin, asin, sinh, asinh, cos, acos, cosh, acosh, tan, atan, tanh, atanh, exp, log, log2, log10, e, pi
• x_min: Start point of the interval within which the function will be evaluated. You can also use constants like pi, e.g. -pi/2
• x_max: End point of the interval within which the function will be evaluated. You can also use constants like pi, e.g. 2*pi
• y_min: Only accessible if mode is set to "Classification". Specify the lower bound of the interval within which samples will be generated
• y_max: Only accessible if mode is set to "Classification". Specify the upper bound of the interval within which samples will be generated

Drawing

You can also draw your own function and use that for the regression or classification task by setting the value of Data source to "Drawing".
You specify everything that is related to this function in the upper right area of the window.

• Main window: In the large area you draw the function. Simply click and keep the mouse button pressed while you move. Release the button to finish the drawing
• x_min: The x-coordinate of the left corners of the drawing area
• x_max: The x-coordinate of the right corners of the drawing area
• y_min: The y-coordinate of the bottom corners of the drawing area
• y_max: The y-coordinate of the upper corners of the drawing area
• Reset: Click this button to delete your drawn function and to draw a different one

Run network

After you have specified all the parameters, you can run the network. Click the button "Run network" at the very bottom of the main window.
Note that the window may seem to be frozen. This does NOT mean that something went wrong. Just wait for a response and don't kill the process. If everything was successful, you will see a visualization of your program output and the ground truth like in the pictures above.
If your program exceeds the time limit or throws and error, you will get notified.
The script also generates a file "input.txt" of the same format that is used on the homework server. This file will be placed in the same directory as the file "main.py"