nn-analysis-gui

Network Analysis GUI using the R network analysis tool from

Getting up and running

Download Docker

• Docker

Docker commands

• docker-compose up
  • This runs the docker-compose.yml file
• docker-compose -f docker-compose.prod.yml up
  • This runs docker compose on the docker-compose.prod.yml up file
• ctrl + c
  • This exits docker
• docker-compose down
  • This stops docker and removes the containers

Start the project

When you run the project for the first time, run

• docker-compose up

Run the project

When you run the project any following time, it will work if you run

• docker-compose up However, if you make any changes to the Dockerfile or requirements.txt, run
• docker-compose up --build to update the Docker image.

File Compatibility

Example network structure .txt files can be found in nn-analysis-gui/testfiles.

File Structure

Consider a network with structure input → h₀ → ... → h_n → output.

The file structure is:
b₀, W₀₀, ..., W_0m, ... b_n, W_n0, ..., W_nm
dim(input), dim(h₀), ..., dim(h_n), dim(output) in_labelh₀, ..., in_labelh_n
out_labelh₀, ..., out_labelh_n

Example

For a 2 → 5 → 5 → 2 network structure with the following attributes,

Weight 1: 
 [[ 2.71078656 -0.54482255  1.49119083  3.40287371  2.72451196]
 [-2.82694977  0.89892445 -1.34958522 -1.37313475 -0.08053674]]
Weight 2: 
 [[-0.71557014  1.93788306  0.53519384 -0.75823443  0.25123773]
 [-0.33080407  2.01800074 -1.42523548 -0.3711385  -2.73905808]
 [-6.10737842  0.32545863  0.74408332 -2.48650943  3.74387307]
 [-3.90689368 -0.86895665 -1.35969632  2.06487152  2.0416786 ]
 [-0.68386149 -0.19280488 -2.77274485 -4.96151007 -1.63668046]]
Weight 3: 
 [[ 0.28411543  0.39069374]
 [-0.29518525 -1.02619704]
 [-0.46183389 -0.17988786]
 [ 0.17719155 -0.15959498]
 [ 1.29132985 -1.11049255]]
Bias 1: 
 [[ 0.11876654 -0.0700461  -0.0576522  -0.13987502 -0.01901942]]
Bias 2: 
 [[ 0.01776611  0.07615346 -0.01073479 -0.00166744  0.06711185]]
Bias 3: 
 [[ 0.10387698 -0.10387698]]

the corresponding text file would be

0.11876654,2.71078656,-2.82694977,-0.0700461,-0.54482255,0.89892445,-0.0576522,1.49119083,-1.34958522,-0.13987502,3.40287371,-1.37313475,-0.01901942,2.72451196,-0.08053674,0.01776611,-0.71557014,-0.33080407,-6.10737842,-3.90689368,-0.68386149,0.07615346,1.93788306,2.01800074,0.32545863,-0.86895665,-0.19280488,-0.01073479,0.53519384,-1.42523548,0.74408332,-1.35969632,-2.77274485,-0.00166744,-0.75823443,-0.3711385,-2.48650943,2.06487152,-4.96151007,0.06711185,0.25123773,-2.73905808,3.74387307,2.0416786,-1.63668046,0.10387698,0.28411543,-0.29518525,-0.46183389,0.17719155,1.29132985,-0.10387698,0.39069374,-1.02619704,-0.17988786,-0.15959498,-1.11049255
2,5,5,2
Mass,Momentum
Electron,Muon

The weights and biases in this file were printed from the weights and biases shown using the following code:

W1, b1, W2, b2,b3,W3 = model['W1'], model['b1'], model['W2'], model['b2'],model['b3'],model["W3"]

for i in range(len(b1[0])):
    print(b1[0][i], end=",")
    for j in range(len(W1)):
        print(W1[j][i], end=",")

for i in range(len(b2[0])):
    print(b2[0][i], end=",")
    for j in range(len(W2)):
        print(W2[j][i], end=",")
        
for i in range(len(b3[0])):
    print(b3[0][i], end=",")
    for j in range(len(W3)):
        print(W3[j][i], end=",")

Interpreting the Results

The output of the analysis will include an illustration of the input neural network model. The output will be in the form shown below. The black lines in the illustration indicate positive weight, the gray lines indicate negative weight, and the thicker the line the greater the magnitude of the weight. The breakdown of the weights in the illustration is as follows:

• 1: Big and positive
• 2: Small and positive
• 3: Big and negative
• 4: Small and negative

Acknowledgements

• This repository was built with subtantial help and guidance from Chad Baily
• The analysis package used in this gui is from Marcus W. Beck and can be found here.