3 # svrt is the ``Synthetic Visual Reasoning Test'', an image
4 # generator for evaluating classification performance of machine
5 # learning systems, humans and primates.
7 # Copyright (c) 2017 Idiap Research Institute, http://www.idiap.ch/
8 # Written by Francois Fleuret <francois.fleuret@idiap.ch>
10 # This file is part of svrt.
12 # svrt is free software: you can redistribute it and/or modify it
13 # under the terms of the GNU General Public License version 3 as
14 # published by the Free Software Foundation.
16 # svrt is distributed in the hope that it will be useful, but
17 # WITHOUT ANY WARRANTY; without even the implied warranty of
18 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 # General Public License for more details.
21 # You should have received a copy of the GNU General Public License
22 # along with selector. If not, see <http://www.gnu.org/licenses/>.
28 from colorama import Fore, Back, Style
34 from torch import optim
35 from torch import FloatTensor as Tensor
36 from torch.autograd import Variable
38 from torch.nn import functional as fn
39 from torchvision import datasets, transforms, utils
43 from vignette_set import VignetteSet, CompressedVignetteSet
45 ######################################################################
47 parser = argparse.ArgumentParser(
48 description = 'Simple convnet test on the SVRT.',
49 formatter_class = argparse.ArgumentDefaultsHelpFormatter
52 parser.add_argument('--nb_train_batches',
53 type = int, default = 1000,
54 help = 'How many samples for train')
56 parser.add_argument('--nb_test_batches',
57 type = int, default = 100,
58 help = 'How many samples for test')
60 parser.add_argument('--nb_epochs',
61 type = int, default = 50,
62 help = 'How many training epochs')
64 parser.add_argument('--batch_size',
65 type = int, default = 100,
66 help = 'Mini-batch size')
68 parser.add_argument('--log_file',
69 type = str, default = 'cnn-svrt.log',
70 help = 'Log file name')
72 parser.add_argument('--compress_vignettes',
73 action='store_true', default = False,
74 help = 'Use lossless compression to reduce the memory footprint')
76 args = parser.parse_args()
78 ######################################################################
80 log_file = open(args.log_file, 'w')
83 print(Fore.RED + 'Logging into ' + args.log_file + Style.RESET_ALL)
89 if pred_log_t is None:
92 elapsed = '+{:.02f}s'.format(t - pred_log_t)
94 s = Fore.BLUE + time.ctime() + ' ' + Fore.GREEN + elapsed + Style.RESET_ALL + ' ' + s
95 log_file.write(s + '\n')
99 ######################################################################
101 # Afroze's ShallowNet
104 # ----------------------
106 # -- conv(21x21 x 6) -> 108x108 6
107 # -- max(2x2) -> 54x54 6
108 # -- conv(19x19 x 16) -> 36x36 16
109 # -- max(2x2) -> 18x18 16
110 # -- conv(18x18 x 120) -> 1x1 120
111 # -- reshape -> 120 1
112 # -- full(120x84) -> 84 1
113 # -- full(84x2) -> 2 1
115 class AfrozeShallowNet(nn.Module):
117 super(AfrozeShallowNet, self).__init__()
118 self.conv1 = nn.Conv2d(1, 6, kernel_size=21)
119 self.conv2 = nn.Conv2d(6, 16, kernel_size=19)
120 self.conv3 = nn.Conv2d(16, 120, kernel_size=18)
121 self.fc1 = nn.Linear(120, 84)
122 self.fc2 = nn.Linear(84, 2)
123 self.name = 'shallownet'
125 def forward(self, x):
126 x = fn.relu(fn.max_pool2d(self.conv1(x), kernel_size=2))
127 x = fn.relu(fn.max_pool2d(self.conv2(x), kernel_size=2))
128 x = fn.relu(self.conv3(x))
130 x = fn.relu(self.fc1(x))
134 ######################################################################
136 def train_model(model, train_set):
137 batch_size = args.batch_size
138 criterion = nn.CrossEntropyLoss()
140 if torch.cuda.is_available():
143 optimizer = optim.SGD(model.parameters(), lr = 1e-2)
145 for e in range(0, args.nb_epochs):
147 for b in range(0, train_set.nb_batches):
148 input, target = train_set.get_batch(b)
149 output = model.forward(Variable(input))
150 loss = criterion(output, Variable(target))
151 acc_loss = acc_loss + loss.data[0]
155 log_string('train_loss {:d} {:f}'.format(e + 1, acc_loss))
159 ######################################################################
161 def nb_errors(model, data_set):
163 for b in range(0, data_set.nb_batches):
164 input, target = data_set.get_batch(b)
165 output = model.forward(Variable(input))
166 wta_prediction = output.data.max(1)[1].view(-1)
168 for i in range(0, data_set.batch_size):
169 if wta_prediction[i] != target[i]:
174 ######################################################################
176 for arg in vars(args):
177 log_string('argument ' + str(arg) + ' ' + str(getattr(args, arg)))
179 ######################################################################
181 for problem_number in range(1, 24):
183 model = AfrozeShallowNet()
185 if torch.cuda.is_available():
188 model_filename = model.name + '_' + \
189 str(problem_number) + '_' + \
190 str(args.nb_train_batches) + '.param'
193 for p in model.parameters(): nb_parameters += p.numel()
194 log_string('nb_parameters {:d}'.format(nb_parameters))
196 need_to_train = False
198 model.load_state_dict(torch.load(model_filename))
199 log_string('loaded_model ' + model_filename)
205 log_string('training_model ' + model_filename)
209 if args.compress_vignettes:
210 train_set = CompressedVignetteSet(problem_number,
211 args.nb_train_batches, args.batch_size,
212 cuda=torch.cuda.is_available())
213 test_set = CompressedVignetteSet(problem_number,
214 args.nb_test_batches, args.batch_size,
215 cuda=torch.cuda.is_available())
217 train_set = VignetteSet(problem_number,
218 args.nb_train_batches, args.batch_size,
219 cuda=torch.cuda.is_available())
220 test_set = VignetteSet(problem_number,
221 args.nb_test_batches, args.batch_size,
222 cuda=torch.cuda.is_available())
224 log_string('data_generation {:0.2f} samples / s'.format(
225 (train_set.nb_samples + test_set.nb_samples) / (time.time() - t))
228 train_model(model, train_set)
229 torch.save(model.state_dict(), model_filename)
230 log_string('saved_model ' + model_filename)
232 nb_train_errors = nb_errors(model, train_set)
234 log_string('train_error {:d} {:.02f}% {:d} {:d}'.format(
236 100 * nb_train_errors / train_set.nb_samples,
238 train_set.nb_samples)
241 nb_test_errors = nb_errors(model, test_set)
243 log_string('test_error {:d} {:.02f}% {:d} {:d}'.format(
245 100 * nb_test_errors / test_set.nb_samples,
250 ######################################################################