MesoNet: weight-history debugging

save weight evolution, not just the last matrix. overly aggressive pruning—check update vs normalization before deciding it's the rule's fault

• reference code: weights/WeightReadout_B2.py in the Diehl–Cook Brian2 port. Compare saved weight matrices from the reference implementation with the custom STDP rule.

XeAe input-to-excitatory weight matrix

• recorded changes: weight-evolution logging and distribution visualizations. Some updates appeared to prune too aggressively; the notebook leaves this as a possible implementation bug rather than a confirmed property of the learning rule.

• preprocessing pointer: numpy.where() for conditional filtering/update of activations. Keep data preprocessing changes separate from plasticity changes when comparing runs.

• review questions: do weights hit their bounds, change sign unexpectedly, or disappear due to the pruning rule? Match each histogram to a training step and sample count. Check whether apparent convergence differences come from readout normalization or initialization rather than the synaptic rule itself.

• References saved in this journal section

• https://www.w3resource.com/python-exercises/numpy/python-numpy-exercise-86.php
• https://github.com/GenevieveF/Diehl-Cook_2015_Brian2/blob/main/weights/WeightReadout_B2.py
• https://brian2.readthedocs.io/en/2.7.1/user/synapses.html
• https://www.geeksforgeeks.org/numpy-where-in-python/
• https://numpy.org/doc/2.1/reference/generated/numpy.where.html