/media/naith/HAlpha/python/generators/venv/bin/python3 /media/naith/HAlpha/02-DP_Self-learning_autonomous_systems_using_deferred_learning/Python/architecture_test_ff_nn.py

=== TEST: STEP CHANGE (Activation: relu) ===

=== COMPARISON OF ACTIVATION FUNCTIONS ON STEP PROBLEM ===

--- Training network with activation function: relu ---

Results for activation function relu:
Training completed in 0.27 seconds (total), 617 epochs
MAE before correction (in region): 0.177778
MAE after correction (in region): 0.003785
MAX deviation before correction (in region): 0.400000
MAX deviation after correction (in region): 0.017283
Error reduction in region: 97.87%
Displaying surface: Corrected Surface - relu activation

--- Training network with activation function: sigmoid ---

Results for activation function sigmoid:
Training completed in 0.01 seconds (total), 21 epochs
MAE before correction (in region): 0.177778
MAE after correction (in region): 0.180051
MAX deviation before correction (in region): 0.400000
MAX deviation after correction (in region): 0.392636
Error reduction in region: -1.28%
Displaying surface: Corrected Surface - sigmoid activation

--- Training network with activation function: tanh ---

Results for activation function tanh:
Training completed in 0.12 seconds (total), 271 epochs
MAE before correction (in region): 0.177778
MAE after correction (in region): 0.081861
MAX deviation before correction (in region): 0.400000
MAX deviation after correction (in region): 0.283176
Error reduction in region: 53.95%
Displaying surface: Corrected Surface - tanh activation
Displaying graph: Activation Function Comparison - STEP

=== SUMMARY OF ACTIVATION FUNCTION RESULTS ===
Activation | Epochs  | Time (s)  | MAE        | MAE region   | MAX error  | Reduction (%)
-------------------------------------------------------------------------------------
relu       | 617     | 0.27      | 0.000324   | 0.003785     | 0.017283   | 97.87
sigmoid    | 21      | 0.01      | 0.015424   | 0.180051     | 0.392636   | -1.28
tanh       | 271     | 0.12      | 0.007012   | 0.081861     | 0.283176   | 53.95
Early stopping - no improvement after 20 epochs
Training completed in 0.03 seconds
Best MAE achieved: 0.002794
Final relative error: 1.57%
MAE before correction (entire surface): 0.015229
MAE after correction (entire surface): 0.000239
MAE before correction (in region): 0.177778
MAE after correction (in region): 0.002794
MAX deviation before correction (entire surface): 0.400000
MAX deviation after correction (entire surface): 0.013890
MAX deviation before correction (in region): 0.400000
MAX deviation after correction (in region): 0.013890
Error reduction in region: 98.43%
Displaying surface: 1) Stavový prostor bez odchylky (original)
Displaying surface: 2) Rozdíl mezi odezvou neuronové sítě a stavovým prostorem bez odchylky
Displaying surface: 3) Stavový prostor s odchylkou (skok)
Displaying surface: 4) Chybový stavový prostor (rozdíl mezi povrchem s chybou a originálem)
Displaying surface: 5) Korigovaný povrch pomocí malé neuronové sítě (aktivace: relu)
Displaying surface: 5) Rozdíl mezi původním a opraveným povrchem (aktivace: relu)

Statistika rozdílu opraveného a původního povrchu:
Průměrná absolutní odchylka: 0.000239
Směrodatná odchylka: 0.001049
Maximální absolutní odchylka: 0.013890

Process finished with exit code 0
