loss_list= []
x =[]
u = 0.001 # константа для скорости обучения

def delta_L(y, x, w, reg = 0.001): #градиент
    return np.sum((-X*(y[:,np.newaxis]))/(1+np.exp(y[:,np.newaxis]*(X.dot(w[:,np.newaxis])))) + reg * w,axis=0)
for epoch in range(10000): #10000 эпох для обучения
    loss = 0

    loss = np.sum(np.log(1+np.exp(-y[:,np.newaxis]*(X.dot(w[:,np.newaxis])))),axis = 0)/len(X)

    w -=  u* delta_L(y,X,w)
    x.append(epoch)
    loss /= len(X)
    loss_list.append(loss)

plt.plot(x,loss_list)