sample_set.py

import os
import random
import torch
import torch.utils.data as data
from pandas.io.parsers import read_csv


def read_files(folder):

    length = 48
    
    daily_data = []

    for root, _, fnames in os.walk(folder):
        for fname in fnames:
            
            path = os.path.join(root, fname)
            df = read_csv(path)
            
            for i in range(0,len(df)):
		if i % length == 0 :
		    if i != 0 :
		        daily_data.append(temp[:])
		    temp = [ df['price_change'][i] ]
		else :
		    temp.append(df['price_change'][i])

    return daily_data


class Sample_set(data.Dataset):

    def __init__(self, folder):

	data = read_files(folder)
	print 'This set contains %d items.' % len(data)
	self.data = data

    def __getitem__(self, index):

	rand_prob = 0.5
        rand_range = 0.1
        
        item_with_noise = self.data[index][:]
        for i in range(0,len(item_with_noise)):
            if random.uniform(0,1) < rand_prob :
                item_with_noise[i] += random.uniform(-rand_range,rand_range)

	item = torch.Tensor(item_with_noise)
        target = torch.Tensor(self.data[index][:])
	return item,target

    def __len__(self):
        return len(self.data)