Optimizing Mind Demo

This is a static demo of a regression network showing the level of interpretability possible with Optimizing Mind.

Inputs

Here are the feature inputs for the regression network.

In [1]:
inputs = [
    [-0.002],
    [ 0.02604],
    [ 0.009076],
    [ 0.011881],
    [-0.010668],
    [ 0.00584 ],
    [-0.007716],
    [ 0.005753],
    [ 0.005881],
]
weights = [
    [-0.02516],
    [ 0.031604],
    [ 0.009076],
    [ 0.011881],
    [-0.010668],
    [ 0.00584 ],
    [-0.007716],
    [ 0.002753],
    [ 0.003881],
]
decision_threshold = 0.7

Results

The explanation magic has already happened elsewhere. Here, you are simply seeing the results.

In [2]:
from IPython.display import Image
Image(filename='saved.png')
Out[2]:

Explanation

Score for part number #9 is 0.47589240202691985.
Network recommends rejection based on threshold 0.726.
With changes to the following factors the network would approve:

#3 NL30CL from 20.00 to 33.16
#4 DIFF10 from -9.80 to -24.46
#2 NL30US from 0.00 to 17.23

3 factors strengthening validation the most:

1) #7 factor EQUITYISEE value 199.00 is above expected by 198.18
2) #3 factor NL30CL value 20.00 is above expected by 16.45
3) #8 factor INDEXISEE value 35.00 is above expected by 33.84

3 factors weakening validation the most:

1) #6 factor ALLSECISEE value 139.00 is below expected by 141.31
2) #2 factor NL30US value 0.00 is below expected by 2.71
3) #1 factor NL5TF value -7.00 is below expected by 7.48

Note that even though the scores of factors EQUITYISEE DIFF10 NL30US ALLSECISEE are within the 3 most different scores from the expected they are not within the 3 most contributing or inhibiting because of model preference.