# HyperNEAT: Weight analysis

## Rotationnaly-invariant
In all these settings, the substrate consists of the input layer with all qubit and plaquette operators, and the output layer is just one node lying at (0,0).

### Example 1
> The thickness of the edges is proportional to the connection weight. Red means negative, blue positive.



> Test syndrome configuration having only one qubit flipped at (2,1) location. The blue node indicates the action taken by the previous NN, it is the node with highest probability (see numbers below)

actions coordinates: [[2, 1], [1, 0], [0, 1], [1, 2]]
corresponding probabilities: [7.193117763274953e-33, 2.30803737252213e-43, 1.773644520392853e-79, 3.8626802679992375e-69]

### Example 2




actions: [[4, 1], [3, 0], [2, 1], [3, 2]]
probs: [0.9636355189868359, 0.9636355189868359, 0.9999999999988003, 0.9636355189868359]

### Example 3




actions: [[4, 1], [3, 0], [2, 1], [3, 2]]
probs: [1.858960384740855e-06, 1.858960384740855e-06, 0.9999732858501016, 1.858960384740855e-06]