#!/usr/bin/python3
# -*- coding: utf-8 -*-
# 　

import numpy as np
import matplotlib.pyplot as plt
import tensorflow as tf
from pandas import DataFrame
from tensorflow.keras import layers, models
import sys
import time
import specrep

t = time.localtime()
print('executed at {:4}/{:02}/{:02} {:02}:{:02}:{:02}'.format(t.tm_year,t.tm_mon,t.tm_mday,t.tm_hour,t.tm_min,t.tm_sec))
print('python version:', sys.version)
print('numpy version:', np.__version__)
print('tensorflow version: ', tf.__version__)
if (not tf.__version__.startswith('2.')):
    print('This code requires version 2.*.')
    exit()


print('\nPrepare the training dateset.')
train_t = np.array([5.2, 5.7, 8.6, 14.9, 18.2, 20.4, 25.5, 26.4, 22.8, 17.5, 11.1, 6.6])
train_t = train_t.reshape([12, 1])
train_x = np.array([[mon**n for n in range(1, 5)] for mon in range(1, 13)])
print('train_x =', train_x)
print('train_t =', train_t)
# x を5次元の入力データとみなし、12組の(x,t)のサンプルと理解する

print('\nDefine a model to predict labels using a polynomal function.')
model = models.Sequential()
model.add(layers.Dense(1, input_shape=(4,), name='polynomial'))
model.summary()

print('\nCompile the model using the Adam optimizer, and MSE (Mean Square Error) as a loss function.')
model.compile(optimizer='adam', loss='mse')

print('\nApply the optimization for 10 epochs.')
history = model.fit(train_x, train_t, batch_size=12, epochs=10)

print('\nApply the optimization for additional 100,000 epochs.')
time_start = time.perf_counter()
history = model.fit(train_x, train_t, batch_size=12, epochs=100000, verbose=0)
time_end = time.perf_counter()
print('time elapsed =', time_end-time_start)

print('\nPlot charts to see how the loss has changed.')
DataFrame({'loss': history.history['loss']}).plot(xlim=(0, 2000))
plt.savefig('ex1-2a.png')
#plt.show()
DataFrame({'loss': history.history['loss']}).plot(ylim=(0, 10))
plt.savefig('ex1-2b.png')
#plt.show()

print('\nShow weights after the training.')
print('weights =', model.get_weights())

print('\nDefine a function to predict values using the final weights.')
w, b = model.get_weights()
def predict(x):
    pred = b[0] + sum([w[n][0] * x**(n+1) for n in range(0, 4)])
    return pred


print('\nPlot a chart for predictions.')
fig = plt.figure()
subplot = fig.add_subplot(1, 1, 1)
subplot.set_xlim(1, 12)
subplot.scatter(range(1, 13), train_t)
xs = np.linspace(1, 12, 100)
ys = predict(xs)
subplot.plot(xs, ys)
plt.savefig('ex1-2.png')
#plt.show()

print('\nLog of computer spec.')
specrep.report()