Updates to several examples

example/reinforcement-learning/ddpg/README.md

@@ -1,6 +1,8 @@
 # mx-DDPG
 MXNet Implementation of DDPG
 
+## /!\ This example depends on RLLAB which is deprecated /!\
+
 # Introduction
 
 This is the MXNet implementation of [DDPG](https://arxiv.org/abs/1509.02971). It is tested in the rllab cart pole environment against rllab's native implementation and achieves comparably similar results. You can substitute with this anywhere you use rllab's DDPG with minor modifications.

example/reinforcement-learning/dqn/setup.sh

@@ -26,11 +26,11 @@ pip install pygame
 
 # Install arcade learning environment
 if [[ "$OSTYPE" == "linux-gnu" ]]; then
-    sudo apt-get install libsdl1.2-dev libsdl-gfx1.2-dev libsdl-image1.2-dev cmake
+    sudo apt-get install libsdl1.2-dev libsdl-gfx1.2-dev libsdl-image1.2-dev cmake ninja-build
 elif [[ "$OSTYPE" == "darwin"* ]]; then
     brew install sdl sdl_image sdl_mixer sdl_ttf portmidi
 fi
-git clone git@github.com:mgbellemare/Arcade-Learning-Environment.git || true
+git clone https://github.com/mgbellemare/Arcade-Learning-Environment || true
 pushd .
 cd Arcade-Learning-Environment
 mkdir -p build
@@ -43,6 +43,6 @@ popd
 cp Arcade-Learning-Environment/ale.cfg .
 
 # Copy roms
-git clone git@github.com:npow/atari.git || true
+git clone https://github.com/npow/atari || true
 cp -R atari/roms .
 

example/restricted-boltzmann-machine/README.md

@@ -8,6 +8,58 @@ Here are some samples generated by the RBM with the default hyperparameters. The
 
 <p style="text-align:center"><img src="samples.png"/></p>
 
+Usage:
+
+```
+python binary_rbm_gluon.py --help
+usage: binary_rbm_gluon.py [-h] [--num-hidden NUM_HIDDEN] [--k K]
+                           [--batch-size BATCH_SIZE] [--num-epoch NUM_EPOCH]
+                           [--learning-rate LEARNING_RATE]
+                           [--momentum MOMENTUM]
+                           [--ais-batch-size AIS_BATCH_SIZE]
+                           [--ais-num-batch AIS_NUM_BATCH]
+                           [--ais-intermediate-steps AIS_INTERMEDIATE_STEPS]
+                           [--ais-burn-in-steps AIS_BURN_IN_STEPS] [--cuda]
+                           [--no-cuda] [--device-id DEVICE_ID]
+                           [--data-loader-num-worker DATA_LOADER_NUM_WORKER]
+
+Restricted Boltzmann machine learning MNIST
+
+optional arguments:
+  -h, --help            show this help message and exit
+  --num-hidden NUM_HIDDEN
+                        number of hidden units
+  --k K                 number of Gibbs sampling steps used in the PCD
+                        algorithm
+  --batch-size BATCH_SIZE
+                        batch size
+  --num-epoch NUM_EPOCH
+                        number of epochs
+  --learning-rate LEARNING_RATE
+                        learning rate for stochastic gradient descent
+  --momentum MOMENTUM   momentum for the stochastic gradient descent
+  --ais-batch-size AIS_BATCH_SIZE
+                        batch size for AIS to estimate the log-likelihood
+  --ais-num-batch AIS_NUM_BATCH
+                        number of batches for AIS to estimate the log-
+                        likelihood
+  --ais-intermediate-steps AIS_INTERMEDIATE_STEPS
+                        number of intermediate distributions for AIS to
+                        estimate the log-likelihood
+  --ais-burn-in-steps AIS_BURN_IN_STEPS
+                        number of burn in steps for each intermediate
+                        distributions of AIS to estimate the log-likelihood
+  --cuda                train on GPU with CUDA
+  --no-cuda             train on CPU
+  --device-id DEVICE_ID
+                        GPU device id
+  --data-loader-num-worker DATA_LOADER_NUM_WORKER
+                        number of multithreading workers for the data loader
+```
+Default:
+```
+Namespace(ais_batch_size=100, ais_burn_in_steps=10, ais_intermediate_steps=10, ais_num_batch=10, batch_size=80, cuda=True, data_loader_num_worker=4, device_id=0, k=30, learning_rate=0.1, momentum=0.3, num_epoch=130, num_hidden=500)
+```
 [1] G E Hinton &amp; R R Salakhutdinov, Reducing the Dimensionality of Data with Neural Networks Science **313**, 5786 (2006)<br/>
 [2] R M Neal, Annealed importance sampling. Stat Comput **11** 2 (2001)<br/>
 [3] R Salakhutdinov &amp; I Murray, On the quantitative analysis of deep belief networks. In Proc. ICML '08 **25** (2008)