What I pre-planned
- atman : sine wave regression improvement
- implement network config generation script
- get out of stuck local minima
What new did I learn
- understood how the architecture affects the output function
- matplotlib is slow for multiprocessor plotting, lots of errors so shifted to pyqt6
- if I seed some good weights and biases with added noise then training is much faster
- multitrain batches of children like genetic algorithms
Where/How did I implement it
- simpler networks with tanh activation works best for sine wave
- pyqt6 is overkill but made code a lot faster to run
- got to 45% accuracy with 15 generations of 10 children each
Any memorable moments of the week
- broke my toenail while playing football
- played some really good badminton