trival

cfgs/cla/mnist_test.json

@@ -0,0 +1,85 @@
+{
+    "config name" : "mnist_classification",
+
+    "dataset" : "mnist",
+    "dataset params" : {
+        "semi-supervised" : true,
+        "labelled indices filepath" : "./mnist_temp_data_dir/method_1_9.pkl",
+        "output shape" : [28, 28, 1],
+        "batch_size" : 128
+    },
+
+    "assets dir" : "assets/mnist/tests",
+
+    "model" : "classification",
+    "model params" : {
+        "name" : "mnist",
+
+        "input shape" : [28, 28, 1],
+        "nb classes" : 10,
+
+        "optimizer" : "adam",
+        "optimizer params" : {
+            "lr" : 0.001,
+            "lr scheme" : "exponential",
+            "lr params" : {
+                "decay_steps" : 10000,
+                "decay_rate" : 0.1
+            },
+            "beta1" : 0.5,
+            "beta2" : 0.9
+        },
+        "classification loss" : "cross entropy",
+
+        "summary" : true,
+
+        "classifier" : "classifier",
+        "classifier params" : {
+            "batch_norm" : "fused_batch_norm",
+
+            "including_conv" : true,
+            "conv_nb_blocks" : 3,
+            "conv_nb_layers" : [2, 2, 2],
+            "conv_nb_filters" : [32, 64, 128],
+            "conv_nb_ksize" : [3, 3, 3],
+            "no_maxpooling" : true,
+
+            "including_top" : true,
+            "fc_nb_nodes" : [600, 600],
+
+            "output_dims" : 10,
+            "output_activation" : "none",
+
+            "debug" : true
+        }
+    },
+
+    "trainer" : "supervised",
+    "trainer params" : {
+
+        "summary hyperparams string" : "method_1_9",
+
+        "continue train" : false,
+        "multi thread" : true,
+        "batch_size" : 32,
+        "train steps" : 20000,
+        "summary steps" : 1000,
+        "log steps" : 100,
+        "save checkpoint steps" : 10000,
+
+        "debug" : true,
+
+        "validators" : [
+            {
+                "validator" : "dataset_validator",
+                "validate steps" : 1000,
+                "has summary" : true,
+                "validator params" : {
+                    "metric" : "accuracy",
+                    "metric type" : "top1"
+                }
+            }
+        ]
+    }
+}
+

dataset/base_simple_dataset.py

@@ -61,7 +61,6 @@ def __init__(self, config):
 		self.y_test = None
 		self.nb_classes = None
 
-
 	def build_dataset(self):
 		assert(self.x_train is not None and self.y_train is not None)
 		assert(self.x_test is not None and self.y_test is not None)
@@ -77,8 +76,7 @@ def build_dataset(self):
 				os.makedirs(self.extra_file_path)
 
 			# if semisupervised training, prepare labelled train set indices,
-			self.nb_labelled_images_per_class = self.config.get('nb_labelled_images_per_class', 100)
-			self.labelled_image_indices = self._get_labelled_image_indices(self.nb_labelled_images_per_class)
+			self.labelled_image_indices = self._get_labelled_image_indices()
 
 			# unlabelled train set
 			self.x_train_u = self.x_train
@@ -93,8 +91,14 @@ def build_dataset(self):
 			self.x_train_u = self.x_train
 
 
-	def _get_labelled_image_indices(self, nb_images_per_class):
-		pickle_filepath = os.path.join(self.extra_file_path, 'labelled_image_indices_%d.pkl'%nb_images_per_class)
+	def _get_labelled_image_indices(self):
+
+		if 'labelled indices filepath' in self.config:
+			pickle_filepath = self.config['labelled indices filepath']
+		else:
+			nb_images_per_class = self.config.get('nb_labelled_images_per_class', 100)
+			pickle_filepath = os.path.join(self.extra_file_path, 'labelled_image_indices_%d.pkl'%nb_images_per_class)
+
 		if os.path.exists(pickle_filepath):
 			return pickle.load(open(pickle_filepath, 'rb'))
 		else:

train_batch.py

@@ -0,0 +1,84 @@
+# -*- coding: utf-8 -*-
+# MIT License
+# 
+# Copyright (c) 2018 ZhicongYan
+#
+# Permission is hereby granted, free of charge, to any person obtaining a copy
+# of this software and associated documentation files (the "Software"), to deal
+# in the Software without restriction, including without limitation the rights
+# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+# copies of the Software, and to permit persons to whom the Software is
+# furnished to do so, subject to the following conditions:
+#
+# The above copyright notice and this permission notice shall be included in all
+# copies or substantial portions of the Software.
+#
+# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+# SOFTWARE.
+# ==============================================================================
+
+import os
+import sys
+import argparse
+import time
+from datetime import datetime
+from shutil import copyfile
+
+import tensorflow as tf
+
+sys.path.append('./')
+sys.path.append('./lib')
+sys.path.append('../')
+
+from cfgs.networkconfig import get_config
+from dataset.dataset import get_dataset
+from model.model import get_model
+from trainer.trainer import get_trainer
+
+parser = argparse.ArgumentParser(description='')
+parser.add_argument('--gpu',     type=str,   default='0')
+parser.add_argument('--config_file',    type=str,   default='cvae1')		# target config file, stored in ./cfgs
+parser.add_argument('--disp_config',    type=bool,   default=False)			# if there is error in config file, set True to print the config file with line number
+
+args = parser.parse_args()
+
+if __name__ == '__main__':
+	os.environ["CUDA_DEVICE_ORDER"]="PCI_BUS_ID"   # see issue #152
+	os.environ["CUDA_VISIBLE_DEVICES"]=args.gpu
+	tf.reset_default_graph()
+
+	# load config file
+	config = get_config(args.config_file, args.disp_config)
+
+	# make the assets directory and copy the config file to it
+	# so if you want to reproduce the result in assets dir
+	# just copy the config_file.json to ./cfgs folder and run python3 train.py --config=(config_file)
+	if not os.path.exists(config['assets dir']):
+		os.makedirs(config['assets dir'])
+	cfg_filename = datetime.now().strftime('config_file_%y-%m-%d_%H-%M-%S.json')
+	copyfile(os.path.join('./cfgs', args.config_file + '.json'), 
+			os.path.join(config['assets dir'], cfg_filename))
+
+	# prepare dataset
+	dataset = get_dataset(config['dataset'], config['dataset params'])
+
+	tfconfig = tf.ConfigProto()
+	tfconfig.gpu_options.allow_growth = True
+
+	with tf.Session(config=tfconfig) as sess:
+
+		# build model
+		config['model params']['assets dir'] = config['assets dir']
+		model = get_model(config['model'], config['model params'])
+
+		# start training
+		config['trainer params']['assets dir'] = config['assets dir']
+		trainer = get_trainer(config['trainer'], config['trainer params'], model, sess)
+
+		trainer.train(sess, dataset, model)
+

validator/base_validator.py

@@ -58,28 +58,31 @@ def validate(self, model, dataset, sess, step):
 	def parallel_data_reading(self, dataset, indices, phase, method, buffer_depth, nb_threads=4):
 
 		self.t_should_stop = False
+
 		data_queue = queue.Queue(maxsize=buffer_depth)
 
-		def read_data_inner_loop(dataset, data_inner_queue, indices, t_ind, nb_threads):
+		def read_data_inner_loop(dataset, data_queue, indices, t_ind, nb_threads):
 			for i, ind in enumerate(indices):
 				if i % nb_threads == t_ind:
 					# read img and label by its index
 					img, label = dataset.read_image_by_index(ind, 'val', 'supervised')
 					if isinstance(img, list) and isinstance(label, list):
 						for _img, _label in zip(img, label):
-							data_inner_queue.put((img, label))
+							data_queue.put((img, label))
 					elif img is not None:
-						data_inner_queue.put((img, label))
+						data_queue.put((img, label))
+
 
-		def read_data_loop(indices, dataset, data_inner_queue, nb_threads):
+		def read_data_loop(indices, dataset, data_queue, nb_threads):
 			threads = [threading.Thread(target=read_data_inner_loop, 
-								args=(dataset, data_inner_queue, indices, t_ind, nb_threads)) for t_ind in range(nb_threads)]
+				args=(dataset, data_queue, indices, t_ind, nb_threads)) for t_ind in range(nb_threads)]
 			for t in threads:
 				t.start()
 			for t in threads:
 				t.join()
 			self.t_should_stop = True
 
+
 		t = threading.Thread(target=read_data_loop, args=(indices, dataset, data_queue, nb_threads))
 		t.start()
 

validator/dataset_validator.py

@@ -100,19 +100,18 @@ def validate(self, model, dataset, sess, step):
 		nb_samples = np.minimum(len(indices), self.nb_samples)
 		indices = np.random.choice(indices, size=nb_samples, replace=False)
 
+
 		self.t_should_stop = False
-		t, data_queue = self.parallel_data_reading(dataset, indices, 'val', 'supervised', self.batch_size(self.buffer_depth))
+		t, data_queue = self.parallel_data_reading(dataset, indices, 'val', 'supervised', self.batch_size*self.buffer_depth)
 
 		batch_x = []
 		batch_y = []
 
-		while not self.t_should_stop:
+		while not self.t_should_stop or not data_queue.empty():
 			if not data_queue.empty():
 				img, label = data_queue.get()
 				batch_x.append(img)
 				batch_y.append(label)
-			else:
-				time.sleep(1)
 
 			if len(batch_x) == self.batch_size:
 				batch_p = model.predict(sess, np.array(batch_x))
@@ -128,6 +127,7 @@ def validate(self, model, dataset, sess, step):
 
 		t.join()
 
+
 		label_list = np.concatenate(label_list, axis=0)
 		pred_list = np.concatenate(pred_list, axis=0)