vignette_set.py

   1
   2 #  svrt is the ``Synthetic Visual Reasoning Test'', an image
   3 #  generator for evaluating classification performance of machine
   4 #  learning systems, humans and primates.
   5 #
   6 #  Copyright (c) 2017 Idiap Research Institute, http://www.idiap.ch/
   7 #  Written by Francois Fleuret <francois.fleuret@idiap.ch>
   8 #
   9 #  This file is part of svrt.
  10 #
  11 #  svrt is free software: you can redistribute it and/or modify it
  12 #  under the terms of the GNU General Public License version 3 as
  13 #  published by the Free Software Foundation.
  14 #
  15 #  svrt is distributed in the hope that it will be useful, but
  16 #  WITHOUT ANY WARRANTY; without even the implied warranty of
  17 #  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  18 #  General Public License for more details.
  19 #
  20 #  You should have received a copy of the GNU General Public License
  21 #  along with selector.  If not, see <http://www.gnu.org/licenses/>.
  22
  23 import torch
  24 from math import sqrt
  25 from torch import multiprocessing
  26
  27 from torch import Tensor
  28 from torch.autograd import Variable
  29
  30 import svrt
  31
  32 ######################################################################
  33
  34 def generate_one_batch(s):
  35     problem_number, batch_size, random_seed = s
  36     svrt.seed(random_seed)
  37     target = torch.LongTensor(batch_size).bernoulli_(0.5)
  38     input = svrt.generate_vignettes(problem_number, target)
  39     input = input.float().view(input.size(0), 1, input.size(1), input.size(2))
  40     return [ input, target ]
  41
  42 class VignetteSet:
  43
  44     def __init__(self, problem_number, nb_batches, batch_size, cuda = False):
  45         self.cuda = cuda
  46         self.batch_size = batch_size
  47         self.problem_number = problem_number
  48         self.nb_batches = nb_batches
  49         self.nb_samples = self.nb_batches * self.batch_size
  50
  51         seeds = torch.LongTensor(self.nb_batches).random_()
  52         mp_args = []
  53         for b in range(0, self.nb_batches):
  54             mp_args.append( [ problem_number, batch_size, seeds[b] ])
  55
  56         self.data = []
  57         for b in range(0, self.nb_batches):
  58             self.data.append(generate_one_batch(mp_args[b]))
  59
  60         # Weird thing going on with the multi-processing, waiting for more info
  61
  62         # pool = multiprocessing.Pool(multiprocessing.cpu_count())
  63         # self.data = pool.map(generate_one_batch, mp_args)
  64
  65         acc = 0.0
  66         acc_sq = 0.0
  67         for b in range(0, self.nb_batches):
  68             input = self.data[b][0]
  69             acc += input.sum() / input.numel()
  70             acc_sq += input.pow(2).sum() /  input.numel()
  71
  72         mean = acc / self.nb_batches
  73         std = sqrt(acc_sq / self.nb_batches - mean * mean)
  74         for b in range(0, self.nb_batches):
  75             self.data[b][0].sub_(mean).div_(std)
  76             if cuda:
  77                 self.data[b][0] = self.data[b][0].cuda()
  78                 self.data[b][1] = self.data[b][1].cuda()
  79
  80     def get_batch(self, b):
  81         return self.data[b]
  82
  83 ######################################################################
  84
  85 class CompressedVignetteSet:
  86     def __init__(self, problem_number, nb_batches, batch_size, cuda = False):
  87         self.cuda = cuda
  88         self.batch_size = batch_size
  89         self.problem_number = problem_number
  90         self.nb_batches = nb_batches
  91         self.nb_samples = self.nb_batches * self.batch_size
  92         self.targets = []
  93         self.input_storages = []
  94
  95         acc = 0.0
  96         acc_sq = 0.0
  97         for b in range(0, self.nb_batches):
  98             target = torch.LongTensor(self.batch_size).bernoulli_(0.5)
  99             input = svrt.generate_vignettes(problem_number, target)
 100             acc += input.float().sum() / input.numel()
 101             acc_sq += input.float().pow(2).sum() /  input.numel()
 102             self.targets.append(target)
 103             self.input_storages.append(svrt.compress(input.storage()))
 104
 105         self.mean = acc / self.nb_batches
 106         self.std = sqrt(acc_sq / self.nb_batches - self.mean * self.mean)
 107
 108     def get_batch(self, b):
 109         input = torch.ByteTensor(svrt.uncompress(self.input_storages[b])).float()
 110         input = input.view(self.batch_size, 1, 128, 128).sub_(self.mean).div_(self.std)
 111         target = self.targets[b]
 112
 113         if self.cuda:
 114             input = input.cuda()
 115             target = target.cuda()
 116
 117         return input, target
 118
 119 ######################################################################