"""
gradient descent for neural networks
"""

# Each loop:
# - computes predictions
# - compute derivatives: dw, db
# update: w = w - alpha * dw; b = b - ....