def fit_and_get_errors(X_train, X_test, y_train, y_test):

    Train_err=[]
    Test_err=[]
    for d in range(1, 21):
        X_tilda,w = get_model(X_train, y_train, d)
        h1, E_train = get_prediction_and_error(X_tilda, y_train, w)
        Train_err.append(np.log(E_train))

        X_temp, w1 = get_model(X_train, y_train, d)
        h2, E_test = get_prediction_and_error(X_temp, y_test, w)
        Test_err.append(np.log(E_test))

    return Train_err, Test_err

#plotanje
fig, axs = plt.subplots(3, 3, figsize=(15, 15))
axs = axs.ravel()

degree=list(range(1, 21))
noise = [100, 200, 500]
# N = [50, 100, 500] #ovliko cemo uzmiat iz test/train skupa
K = [1/3, 2/3, 1]

X = make_instances(-5,5,1000)
X_train, X_test = train_test_split(X, test_size=0.5)


for i, sigma in enumerate(noise):
    y = list(make_labels(X_train, f, sigma))
    y_train = np.array(y)
    y = list(make_labels(X_test,f,sigma))
    y_test = np.array(y)

    for j, n in enumerate(K):
        number_of_examples=math.floor(n*500)
        index_list = np.random.choice(np.arange(500), number_of_examples, replace=False)
        X1_train = X_train[index_list]
        y1_train = y_train[index_list]


#         index_list = np.random.choice(np.arange(500), number_of_examples, replace=False)
        X1_test = X_test[index_list]
        y1_test = y_test[index_list]

        #dobivanje train i test erora
        train,test=fit_and_get_errors(X1_train,X1_test,y1_train,y1_test)

        axs[i*3+j].set_title("N = " + str(math.ceil(n*500)) + ", noise = " + str(sigma))
        axs[i*3+j].plot(degree,train,label="Train")
        axs[i*3+j].plot(degree,test,label="Test")
        axs[i*3+j].legend()


plt.show()