robustness

README.md

@@ -12,10 +12,9 @@
     - there is also a distance matrix for all-to-all particle comparisons (with distance parameter one of: `MSE`, `MAE` (mean absolute error = mean manhattan) and `MIM` (mean position invariant manhattan))
 
 
-- [ ] Same Thing with Soup interactionWe would expect the same behaviour...Influence of interaction with near and far away particles.
-
-- [ ] How are basins / "attractor areas" shaped?
-      - Weired.... tbc...
+- [ ] Same Thing with Soup interaction. We would expect the same behaviour...Influence of interaction with near and far away particles.
+      - 
+      - 
   
 - [x] Robustness test with a trained NetworkTraining for high quality fixpoints, compare with the "perfect" fixpoint. Average Loss per application step
 
@@ -25,17 +24,23 @@
 
     - We might need to consult about the "average loss per application step", as I think application loss get gradually higher the worse the weights get. So the average might not tell us much here.
 
-- [ ] Adjust Self Training so that it favors second order fixpoints-> Second order test implementation (?)
+- [x] Adjust Self Training so that it favors second order fixpoints-> Second order test implementation (?)
 
 - [x] Barplot over clones -> how many become a fixpoint cs how many diverge per noise level
 
-- [ ] Box-Plot of Avg. Distance of clones from parent
+- [x] Box-Plot of Avg. Distance of clones from parent
+
+- [ ] Search subspace between two fixpoints by linage(10**-5), check were they end up
+
+- [ ] How are basins / "attractor areas" shaped?
+      - Weired.... tbc...
+      - 
+
 
 # Future Todos:
 
 - [ ] Find a statistik over weight space that provides a better init function
 - [ ] Test this init function on a mnist classifier - just for the lolz
-- [ ] 
 
 ---
 ## Notes: 

functionalities_test.py

@@ -24,9 +24,11 @@ def is_identity_function(network: Net, epsilon=pow(10, -5)) -> bool:
                        rtol=0, atol=epsilon)
 
 
-def is_zero_fixpoint(network: Net) -> bool:    
-    result = bool(len(np.nonzero(network.create_target_weights(network.input_weight_matrix()))))
-    return not result
+def is_zero_fixpoint(network: Net, epsilon=pow(10, -5)) -> bool:
+    target_data = network.create_target_weights(network.input_weight_matrix().detach().numpy())
+    result = np.allclose(target_data, np.zeros_like(target_data), rtol=0, atol=epsilon)
+    # result = bool(len(np.nonzero(network.create_target_weights(network.input_weight_matrix()))))
+    return result
 
 
 def is_secondary_fixpoint(network: Net, epsilon: float = pow(10, -5)) -> bool:

journal_robustness.py

@@ -225,7 +225,7 @@ if __name__ == "__main__":
     ST_net_hidden_size = 2
     ST_net_learning_rate = 0.04
     ST_name_hash = random.getrandbits(32)
-    ST_synthetic = True
+    ST_synthetic = False
 
     print(f"Running the robustness comparison experiment:")
     RobustnessComparisonExperiment(