# -*- coding=utf-8 -*-
import numpy as np
import tensorflow as tf
from sklearn.metrics import mean_squared_error, r2_score


def normalize(x):
    mean = np.mean(x)
    std = np.std(x)
    return (x - mean) / std


def append_bias_reshape(features, labels):
    m = features.shape[0]
    n = features.shape[1]
    x = np.reshape(np.c_[np.ones(m), features], [m, n + 1])
    y = np.reshape(labels, [m, 1])
    return x, y


def load_data():
    p = r'/Users/light/workspace/courses/numerical_analysis/8/wine.csv'
    with open(p, encoding='utf-8') as f:
        data = np.loadtxt(f, delimiter=";", skiprows=1).astype('float64')
        return data[:, :11], data[:, 11:]


x_all, y_all = load_data()
x_test = x_all[1400:, :11]
y_test = y_all[1400:]
y_train = y_all[:1400]
x_train = normalize(x_all[:1400, 0:1])
for i in range(1, 11):
    x_train = np.concatenate((x_train, normalize(x_all[:1400, i:i + 1])), 1)

# x_train, y_train = append_bias_reshape(x_train, y_train)
# dataset = tfdata.Dataset.from_tensor_slices((x_train, y_train))

print(x_train.dtype)
W = tf.Variable(np.random.random([11, 1]))
b = tf.Variable(np.random.random([1, 1]))


def linear_regression(x):
    return tf.matmul(x, W) + b


# 均方差
def mean_square(y_pred, y_true):
    return tf.reduce_sum(tf.pow(y_pred - y_true, 2)) / (2 * 1400)


optimizer = tf.optimizers.SGD(0.01)


# 优化过程
def run_optimization():
    # 将计算封装在GradientTape中以实现自动微分
    with tf.GradientTape() as g:
        pred = linear_regression(x_train)
        loss = mean_square(pred, y_train)

    # 计算梯度
    gradients = g.gradient(loss, [W, b])

    # 按gradients更新 W 和 b
    optimizer.apply_gradients(zip(gradients, [W, b]))


training_steps = 1000
display_step = 50

# 针对给定训练步骤数开始训练
for step in range(1, training_steps + 1):
    # 运行优化以更新W和b值
    run_optimization()

    if step % display_step == 0:
        pred = linear_regression(x_train)
        loss = mean_square(pred, y_train)
        print("step: %i, loss: %f, W: %s, b: %s" % (step, loss, W.numpy(), b.numpy()))

result = linear_regression(x_test).numpy().round()
delta = result - y_test
print(f"根均方误差(RMSE)：{np.sqrt(mean_squared_error(result, y_test))}")
print(f"测试集R^2：{r2_score(y_test, result)}")