import warnings
warnings.filterwarnings("ignore", message="numpy.dtype size changed")

import numpy as np
import pandas as pd


def load_data():
	#df = pd.read_csv('data/sample.txt', delimiter=',', header=None)
	#df[2] = pd.Series([1]*len(df[0]))

	#df = pd.read_csv('data/ex1data1.txt', delimiter=',', header=None)
	#df[2] = pd.Series([1]*len(df[0]))

	df = pd.read_csv('data/train.csv', delimiter=',', comment='#')
	df[0] = df['x']
	df[1] = df['y']

	df[2] = pd.Series([1]*len(df[0]))
	df = df.reindex(columns=[1, 2, 0])

	return df

def feature_scaling(x):
	mean = x.mean(0)
	mean[0] = 0
	std = x.std(0)
	std[std==0] = 1
	std[0] = 1
	x = (x-mean)/std
	return x, mean, std


df = load_data()
y = df.values[:, 0]
x = df.values[:, 1:]
m, n = x.shape
w = np.zeros(n)
a = 1.0e-3

x, mean, std = feature_scaling(x)

sum_ = 0
gap = 0.001
break_gap = 1.0e-10
trycnt = 0
while True:
	h = x@w
	cost = ((h-y)**2).mean()/2
	gradient = a*(h-y)@x/m
	w -= gradient

	sum_ += abs(gradient.sum())
	if sum_ >= gap:
		print('[{}] {} ({})'.format(trycnt, cost, w))
		sum_ -= gap

	trycnt += 1
	if abs(gradient.max()) < break_gap:
		break

	if not np.isfinite(cost) or not np.isfinite(sum_):
		break

print('[{}] {} ({})'.format(trycnt, cost, w))
#print('x: {}'.format(x[:10]))
#print('y: {}'.format(y[:10]))
#print('h: {}'.format(h[:10]))
print(np.c_[x[:10], y[:10], h[:10]])
pass