Fix.

[pysvrt.git] / cnn-svrt.py

diff --git a/cnn-svrt.py b/cnn-svrt.py
index 26e7de8..f3d350e 100755 (executable)
--- a/cnn-svrt.py
+++ b/cnn-svrt.py
@@ -56,6 +56,13 @@ parser.add_argument('--nb_train_samples',
 parser.add_argument('--nb_test_samples',
                     type = int, default = 10000)
 
+parser.add_argument('--nb_validation_samples',
+                    type = int, default = 10000)
+
+parser.add_argument('--validation_error_threshold',
+                    type = float, default = 0.0,
+                    help = 'Early training termination criterion')
+
 parser.add_argument('--nb_epochs',
                     type = int, default = 50)
 
@@ -77,11 +84,15 @@ parser.add_argument('--test_loaded_models',
                     type = distutils.util.strtobool, default = 'False',
                     help = 'Should we compute the test errors of loaded models')
 
+parser.add_argument('--problems',
+                    type = str, default = '1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23',
+                    help = 'What problems to process')
+
 args = parser.parse_args()
 
 ######################################################################
 
-log_file = open(args.log_file, 'w')
+log_file = open(args.log_file, 'a')
 pred_log_t = None
 
 print(Fore.RED + 'Logging into ' + args.log_file + Style.RESET_ALL)
@@ -190,7 +201,22 @@ class AfrozeDeepNet(nn.Module):
 
 ######################################################################
 
-def train_model(model, train_set):
+def nb_errors(model, data_set):
+    ne = 0
+    for b in range(0, data_set.nb_batches):
+        input, target = data_set.get_batch(b)
+        output = model.forward(Variable(input))
+        wta_prediction = output.data.max(1)[1].view(-1)
+
+        for i in range(0, data_set.batch_size):
+            if wta_prediction[i] != target[i]:
+                ne = ne + 1
+
+    return ne
+
+######################################################################
+
+def train_model(model, train_set, validation_set):
     batch_size = args.batch_size
     criterion = nn.CrossEntropyLoss()
 
@@ -212,25 +238,24 @@ def train_model(model, train_set):
             loss.backward()
             optimizer.step()
         dt = (time.time() - start_t) / (e + 1)
+
         log_string('train_loss {:d} {:f}'.format(e + 1, acc_loss),
                    ' [ETA ' + time.ctime(time.time() + dt * (args.nb_epochs - e)) + ']')
 
-    return model
-
-######################################################################
+        if validation_set is not None:
+            nb_validation_errors = nb_errors(model, validation_set)
 
-def nb_errors(model, data_set):
-    ne = 0
-    for b in range(0, data_set.nb_batches):
-        input, target = data_set.get_batch(b)
-        output = model.forward(Variable(input))
-        wta_prediction = output.data.max(1)[1].view(-1)
+            log_string('validation_error {:.02f}% {:d} {:d}'.format(
+                100 * nb_validation_errors / validation_set.nb_samples,
+                nb_validation_errors,
+                validation_set.nb_samples)
+            )
 
-        for i in range(0, data_set.batch_size):
-            if wta_prediction[i] != target[i]:
-                ne = ne + 1
+            if nb_validation_errors / validation_set.nb_samples <= args.validation_error_threshold:
+                log_string('below validation_error_threshold')
+                break
 
-    return ne
+    return model
 
 ######################################################################
 
@@ -250,17 +275,18 @@ def int_to_suffix(n):
 class vignette_logger():
     def __init__(self, delay_min = 60):
         self.start_t = time.time()
+        self.last_t = self.start_t
         self.delay_min = delay_min
 
     def __call__(self, n, m):
         t = time.time()
-        if t > self.start_t + self.delay_min:
+        if t > self.last_t + self.delay_min:
             dt = (t - self.start_t) / m
             log_string('sample_generation {:d} / {:d}'.format(
                 m,
                 n), ' [ETA ' + time.ctime(time.time() + dt * (n - m)) + ']'
             )
-            self.start_t = t
+            self.last_t = t
 
 ######################################################################
 
@@ -268,6 +294,8 @@ if args.nb_train_samples%args.batch_size > 0 or args.nb_test_samples%args.batch_
     print('The number of samples must be a multiple of the batch size.')
     raise
 
+log_string('############### start ###############')
+
 if args.compress_vignettes:
     log_string('using_compressed_vignettes')
     VignetteSet = svrtset.CompressedVignetteSet
@@ -275,7 +303,7 @@ else:
     log_string('using_uncompressed_vignettes')
     VignetteSet = svrtset.VignetteSet
 
-for problem_number in range(1, 24):
+for problem_number in map(int, args.problems.split(',')):
 
     log_string('############### problem ' + str(problem_number) + ' ###############')
 
@@ -322,7 +350,15 @@ for problem_number in range(1, 24):
             train_set.nb_samples / (time.time() - t))
         )
 
-        train_model(model, train_set)
+        if args.validation_error_threshold > 0.0:
+            validation_set = VignetteSet(problem_number,
+                                         args.nb_validation_samples, args.batch_size,
+                                         cuda = torch.cuda.is_available(),
+                                         logger = vignette_logger())
+        else:
+            validation_set = None
+
+        train_model(model, train_set, validation_set)
         torch.save(model.state_dict(), model_filename)
         log_string('saved_model ' + model_filename)
 
@@ -346,10 +382,6 @@ for problem_number in range(1, 24):
                                args.nb_test_samples, args.batch_size,
                                cuda = torch.cuda.is_available())
 
-        log_string('data_generation {:0.2f} samples / s'.format(
-            test_set.nb_samples / (time.time() - t))
-        )
-
         nb_test_errors = nb_errors(model, test_set)
 
         log_string('test_error {:d} {:.02f}% {:d} {:d}'.format(