# -*- coding: utf-8 -*-
"""
Created on Sun Jul  7 15:02:20 2019
"""

import pandas as pd
import numpy as np
import tensorflow as tf
import matplotlib.pyplot as plt

# ---------------
# Prepation of Data
# ---------------

# Import in the required files
xexxar = pd.read_csv("Xexxar.csv")

# We want to analyze Speed and Speed Rating since they are likely to be
# linearly related.
# It is grouped by Speed (Index) and Aggregated with .mean()
xexxar_filtered = xexxar[['Speed', 'Speed Rating']].groupby("Speed").mean()

# Extract x and y (x in nested in the index)
data_x = np.array(xexxar_filtered.index.tolist(), dtype = 'float')
data_y = xexxar_filtered['Speed Rating']

# ---------------
# TensorFlow Preparation
# ---------------

# Define Feeding placeholders, placeholders are slots where data is fed during
# the training process.
feed_x = tf.placeholder(tf.float32)
feed_y = tf.placeholder(tf.float32)

# We define the model as a linear relationship where, y = f(x) = w * x
def model(x, w):
    return(tf.multiply(x, w))

# Initialize w as 1.0, named weight
var_w = tf.Variable(1.0, 'weight')

# Create model with placeholder x and weight variable
model_y = model(feed_x, var_w)

# Define the cost. Square is required to make sure the value is always positive
cost = tf.square(model_y - feed_y)

# We define the optimizer (ADAM) with learning rate = 0.001 and the cost.
cost_op = tf.train.AdamOptimizer(0.001).minimize(cost)

# These are Tensorboard writers

# This defines var_w (weight) to be tracked
summary_w = tf.summary.scalar("Weight", var_w)
# This defines the Tensorboard summary writer
summary_writer = tf.summary.FileWriter("logs_speed")
# This merges all summaries (Not required since there's only 1 summary
# excluding summary_writer)
summary_merged = tf.summary.merge_all()

# Automatically initalize all variables
init = tf.global_variables_initializer()

# Define number of epochs
epochs = 50

# ---------------
# TensorFlow Session
# ---------------

with tf.Session() as sess:
    # Initialize all variables
    sess.run(init)

    # Step is used for Tensorboard Summary
    step = 0
    for epoch in range(epochs):
        for x, y in zip(data_x, data_y):
            step += 1

            # We extract summary_str from the sess.run
            # summary_merged outputs to summary_str
            # cost_op outputs to _ (placeholder)
            summary_str, _ = sess.run([summary_merged, cost_op],
                                      feed_dict={
                                              feed_x: x,
                                              feed_y: y
                                              })

            # Append to Tensorboard Summary
            summary_writer.add_summary(summary_str, step)

        print("Epoch {} \t Weight {}".format(epoch, var_w.eval(sess)))

    # Extract calculated w
    trained_w = var_w.eval(sess)

# ---------------
# Evaluate using matplotlib.pyplot
# ---------------

# Get new trained_y with new trained_y
trained_y = trained_w * data_x

# Plot original
plt.scatter(data_x, data_y)
# Plot new
plt.scatter(data_x, trained_y, c = 'r')

# Show
plt.show()

# Clears Graph when sessions don't close properly sometimes
tf.reset_default_graph()