update example notes

README.md

@@ -9,7 +9,7 @@ You can train them and reproduce the performance... not just to see how to write
 + [InceptionV3 on ImageNet](examples/Inception/inceptionv3.py)
 + [Fully-convolutional Network for Holistically-Nested Edge Detection(HED)](examples/HED)
 + [Spatial Transformer Network on MNIST addition](examples/SpatialTransformer)
-+ [Generative Adversarial Network(GAN) variants](examples/GAN)
++ [Generative Adversarial Network(GAN) variants (DCGAN,Image2Image,InfoGAN)](examples/GAN)
 + [Deep Q-Network(DQN) variants on Atari games](examples/Atari2600)
 + [Asynchronous Advantage Actor-Critic(A3C) with demos on OpenAI Gym](examples/OpenAIGym)
 + [char-rnn language model](examples/char-rnn)

examples/README.md

@@ -3,14 +3,23 @@
 
 Training examples with __reproducible__ and meaningful performance.
 
+## Vision:
 + [An illustrative mnist example with explanation of the framework](mnist-convnet.py)
 + [A tiny SVHN ConvNet with 97.8% accuracy](svhn-digit-convnet.py)
-+ [Inception-BN with 71% accuracy](Inception/inception-bn.py)
-+ [InceptionV3 with 74% accuracy (similar to the official code)](Inception/inceptionv3.py)
 + [DoReFa-Net: binary / low-bitwidth CNN on ImageNet](DoReFa-Net)
 + [ResNet for ImageNet/Cifar10/SVHN](ResNet)
-+ [Holistically-Nested Edge Detection](HED)
++ [Inception-BN with 71% accuracy](Inception/inception-bn.py)
++ [InceptionV3 with 74% accuracy (similar to the official code)](Inception/inceptionv3.py)
++ [Fully-convolutional Network for Holistically-Nested Edge Detection(HED)](HED)
 + [Spatial Transformer Networks on MNIST addition](SpatialTransformer)
-+ [Generative Adversarial Networks variants](GAN)
-+ Reinforcement learning (DQN, A3C) on [Atari games](Atari2600) and [demos on OpenAI Gym](OpenAIGym).
++ Load a pretrained [AlexNet](load-alexnet.py) or [VGG16](load-vgg16.py) model.
+
+## Reinforcement Learning:
++ [Deep Q-Network(DQN) variants on Atari games](Atari2600)
++ [Asynchronous Advantage Actor-Critic(A3C) with demos on OpenAI Gym](OpenAIGym)
+
+## Unsupervised:
++ [Generative Adversarial Network(GAN) variants (DCGAN,Image2Image,InfoGAN)](examples/GAN)
+
+## Speech / NLP:
 + [char-rnn for fun](char-rnn)

examples/ResNet/cifar10-resnet.py

@@ -22,7 +22,7 @@ Identity Mappings in Deep Residual Networks, arxiv:1603.05027
 
 I can reproduce the results on 2 TitanX for
 n=5, about 7.1% val error after 67k steps (8.6 step/s)
-n=18, about 5.9% val error after 80k steps (2.6 step/s)
+n=18, about 5.95% val error after 80k steps (2.6 step/s)
 n=30: a 182-layer network, about 5.6% val error after 51k steps (1.55 step/s)
 This model uses the whole training set instead of a train-val split.
 """

tensorpack/callbacks/param.py

@@ -258,5 +258,6 @@ class StatMonitorParamSetter(HyperParamSetter):
             if hist_max > hist_first + self.threshold: # large enough
                 return None
         self.last_changed_epoch = self.epoch_num
+        logger.info("[StatMonitorParamSetter] Triggered, history: " + ','.join(hist))
         return self.value_func(self.get_current_value())