Replace the numbers of samples by numbers of batches of samples.

[pysvrt.git] / cnn-svrt.py

diff --git a/cnn-svrt.py b/cnn-svrt.py
index 3e20f8e..ab1b363 100755 (executable)
--- a/cnn-svrt.py
+++ b/cnn-svrt.py
@@ -44,18 +44,22 @@ parser = argparse.ArgumentParser(
     formatter_class = argparse.ArgumentDefaultsHelpFormatter
 )
 
-parser.add_argument('--nb_train_samples',
-                    type = int, default = 100000,
+parser.add_argument('--nb_train_batches',
+                    type = int, default = 1000,
                     help = 'How many samples for train')
 
-parser.add_argument('--nb_test_samples',
-                    type = int, default = 10000,
+parser.add_argument('--nb_test_batches',
+                    type = int, default = 100,
                     help = 'How many samples for test')
 
 parser.add_argument('--nb_epochs',
                     type = int, default = 50,
                     help = 'How many training epochs')
 
+parser.add_argument('--batch_size',
+                    type = int, default = 100,
+                    help = 'Mini-batch size')
+
 parser.add_argument('--log_file',
                     type = str, default = 'cnn-svrt.log',
                     help = 'Log file name')
@@ -81,7 +85,7 @@ def generate_set(p, n):
     t = time.time()
     input = svrt.generate_vignettes(p, target)
     t = time.time() - t
-    log_string('DATA_SET_GENERATION {:.02f} sample/s'.format(n / t))
+    log_string('data_set_generation {:.02f} sample/s'.format(n / t))
     input = input.view(input.size(0), 1, input.size(1), input.size(2)).float()
     return Variable(input), Variable(target)
 
@@ -101,9 +105,9 @@ def generate_set(p, n):
 # -- full(120x84)      -> 84         1
 # -- full(84x2)        -> 2          1
 
-class Net(nn.Module):
+class AfrozeShallowNet(nn.Module):
     def __init__(self):
-        super(Net, self).__init__()
+        super(AfrozeShallowNet, self).__init__()
         self.conv1 = nn.Conv2d(1, 6, kernel_size=21)
         self.conv2 = nn.Conv2d(6, 16, kernel_size=19)
         self.conv3 = nn.Conv2d(16, 120, kernel_size=18)
@@ -119,19 +123,14 @@ class Net(nn.Module):
         x = self.fc2(x)
         return x
 
-def train_model(train_input, train_target):
-    model, criterion = Net(), nn.CrossEntropyLoss()
-
-    nb_parameters = 0
-    for p in model.parameters():
-        nb_parameters += p.numel()
-    log_string('NB_PARAMETERS {:d}'.format(nb_parameters))
+def train_model(model, train_input, train_target):
+    bs = args.batch_size
+    criterion = nn.CrossEntropyLoss()
 
     if torch.cuda.is_available():
-        model.cuda()
         criterion.cuda()
 
-    optimizer, bs = optim.SGD(model.parameters(), lr = 1e-2), 100
+    optimizer = optim.SGD(model.parameters(), lr = 1e-2)
 
     for k in range(0, args.nb_epochs):
         acc_loss = 0.0
@@ -142,15 +141,16 @@ def train_model(train_input, train_target):
             model.zero_grad()
             loss.backward()
             optimizer.step()
-        log_string('TRAIN_LOSS {:d} {:f}'.format(k, acc_loss))
+        log_string('train_loss {:d} {:f}'.format(k, acc_loss))
 
     return model
 
 ######################################################################
 
-def nb_errors(model, data_input, data_target, bs = 100):
-    ne = 0
+def nb_errors(model, data_input, data_target):
+    bs = args.batch_size
 
+    ne = 0
     for b in range(0, data_input.size(0), bs):
         output = model.forward(data_input.narrow(0, b, bs))
         wta_prediction = output.data.max(1)[1].view(-1)
@@ -164,25 +164,43 @@ def nb_errors(model, data_input, data_target, bs = 100):
 ######################################################################
 
 for arg in vars(args):
-    log_string('ARGUMENT ' + str(arg) + ' ' + str(getattr(args, arg)))
+    log_string('argument ' + str(arg) + ' ' + str(getattr(args, arg)))
 
 for problem_number in range(1, 24):
-    train_input, train_target = generate_set(problem_number, args.nb_train_samples)
-    test_input, test_target = generate_set(problem_number, args.nb_test_samples)
+    train_input, train_target = generate_set(problem_number,
+                                             args.nb_train_batches * args.batch_size)
+    test_input, test_target = generate_set(problem_number,
+                                           args.nb_test_batches * args.batch_size)
+    model = AfrozeShallowNet()
 
     if torch.cuda.is_available():
         train_input, train_target = train_input.cuda(), train_target.cuda()
         test_input, test_target = test_input.cuda(), test_target.cuda()
+        model.cuda()
 
     mu, std = train_input.data.mean(), train_input.data.std()
     train_input.data.sub_(mu).div_(std)
     test_input.data.sub_(mu).div_(std)
 
-    model = train_model(train_input, train_target)
+    nb_parameters = 0
+    for p in model.parameters():
+        nb_parameters += p.numel()
+    log_string('nb_parameters {:d}'.format(nb_parameters))
+
+    model_filename = 'model_' + str(problem_number) + '.param'
+
+    try:
+        model.load_state_dict(torch.load(model_filename))
+        log_string('loaded_model ' + model_filename)
+    except:
+        log_string('training_model')
+        train_model(model, train_input, train_target)
+        torch.save(model.state_dict(), model_filename)
+        log_string('saved_model ' + model_filename)
 
     nb_train_errors = nb_errors(model, train_input, train_target)
 
-    log_string('TRAIN_ERROR {:d} {:.02f}% {:d} {:d}'.format(
+    log_string('train_error {:d} {:.02f}% {:d} {:d}'.format(
         problem_number,
         100 * nb_train_errors / train_input.size(0),
         nb_train_errors,
@@ -191,7 +209,7 @@ for problem_number in range(1, 24):
 
     nb_test_errors = nb_errors(model, test_input, test_target)
 
-    log_string('TEST_ERROR {:d} {:.02f}% {:d} {:d}'.format(
+    log_string('test_error {:d} {:.02f}% {:d} {:d}'.format(
         problem_number,
         100 * nb_test_errors / test_input.size(0),
         nb_test_errors,