neural network from scratch

nn.py

import numpy as np

def sigmoid( z):
        #Apply sigmoid activation function to scalar, vector, or matrix
        return 1/(1+np.exp(-z))

def sigmoidPrime(z):
        #Gradient of sigmoid
        return np.exp(-z)/((1+np.exp(-z))**2)
        
X = np.array(([3,5], [5,1], [10,2]), dtype=float)
y = np.array(([75], [82], [93]), dtype=float)

# Normalize
X = X/np.amax(X, axis=0)
y = y/100 #Max test score is 100

#input output parameters
inputLayerSize = 2
outputLayerSize = 1
hiddenLayerSize = 3

#random weights for input to hidden
W1 = np.random.randn(inputLayerSize,hiddenLayerSize)

#random weights for output to hidden
W2 = np.random.randn(hiddenLayerSize,outputLayerSize)

#forward prpogation
z2 = np.dot(X, W1)
a2 = sigmoid(z2) #input to hidden activation function
z3 = np.dot(a2, W2)
yHat = sigmoid(z3) #hidden to output activation function

#cost function calculating for layers
delta3 = np.multiply(-(y-yHat), sigmoidPrime(z3)) #output to hidden layer
dJdW2 = np.dot(a2.T, delta3)

delta2 = np.dot(delta3, W2.T)*sigmoidPrime(z2) #hidden to input layer
dJdW1 = np.dot(X.T, delta2)