Building Neural Networks from Scratch -Lecture 9 From Output Error to Weight Updates скачать в хорошем качестве

Building Neural Networks from Scratch -Lecture 9 From Output Error to Weight Updates

In this video, I break down how errors are calculated at each neuron during backpropagation. I explain how the output error is first computed and then propagated backward through the network. import numpy as np class NN: def __init__(self,no_neurons_IL, no_neurons_L1, no_neurons_l2): self.no_neurons_IL = no_neurons_IL self.no_neurons_L1 = no_neurons_L1 self.no_neurons_l2 = no_neurons_l2 self.stage_one_weights = np.random.rand(self.no_neurons_L1,self.no_neurons_IL) self.stage_two_weights = np.random.rand(self.no_neurons_l2,self.no_neurons_L1) def sigmoid (x): return 1/(1+np.exp(-x)) loan_pred = NN(2,3,1) #print(loan_pred.stage_one_weights) #print("==========================================") #print(loan_pred.stage_two_weights) #print("==========================================") input_n1_n2 = np.array([[0.1],[0.6]]) input_to_L1 =(np.matmul(loan_pred.stage_one_weights,input_n1_n2)) #print(input_to_L1) out_from_L1= sigmoid(input_to_L1) #print(out_from_L1) input_to_L2 = (np.matmul(loan_pred.stage_two_weights,input_to_L1)) out_from_L2= sigmoid(input_to_L2) #print(out_from_L2) target= [[1]] En6= target - out_from_L2 #print(En6) En3_En4_En6= np.matmul(loan_pred.stage_two_weights.T,En6) print(En3_En4_En6)