fix documents

Author: Yusuke Oda

doc/decoding.md

@@ -2,23 +2,23 @@ Generating translations using the trained model
 ===============================================
 
 If we obtained the `best_dev_log_ppl.model.params` file in the output directory
-by [training process](https://github.com/odashi/nmtkit/tree/master/doc/training_ja.md),
+by [training process](https://github.com/odashi/nmtkit/tree/master/doc/training.md),
 we can generate output sentences using `decode` command:
 
-    src/bin/decode \
+    $ path/to/decode \
         --model model \
-        < submodules/small_parallel_enja/test.en \
+        < /path/to/nmtkit/submodules/small_parallel_enja/test.en \
         > result.ja
 
 Note that the input data of the `decode` should be tokenized in advance.
 
 If you want to generate HTML file with more detailed information during the
 decoding process, use `--format html` option:
 
-    src/bin/decode \
+    $ /path/to/decode \
         --model model \
         --format html \
-        < submodules/small_parallel_enja/test.en \
+        < /path/to/nmtkit/submodules/small_parallel_enja/test.en \
         > result.ja.html
 
 Here is the

doc/install.md

@@ -9,17 +9,17 @@ NMTKit needs following libraries:
 
 * **GNU C++** ... v4.9 or later (or other compatible compilers)
 * **Boost C++ Library** ... v1.50 or later
-* **Eigen** ... The newest development version
-* **DyNet** ... v1.0-rc1.
+* **[Eigen](http://eigen.tuxfamily.org/)** ... The newest development version
+* **[DyNet](https://github.com/clab/dynet)** ... v1.0-rc1 or later
+* **[MTEval](https://github.com/odashi/mteval)** ... v1.0.0 or later
 * **CUDA** ... v7.5 or later
 
 
 And the installation process may require following tools:
 
 * **Git**
 * **Mercurial**
-* **autotools (autoconf, automake), libtool**
-* **CMake**.
+* **CMake** ... v3.1 or later
 
 
 Install Eigen
@@ -29,67 +29,95 @@ First we get the development version of Eigen using Mercurial.
 This process could be done by only putting obtained files into an appropriate
 location because Eigen consists of only header files:
 
-    hg clone https://bitbucket.org/eigen/eigen/ /path/to/eigen
+    $ hg clone https://bitbucket.org/eigen/eigen/ /path/to/eigen
 
 
 Install DyNet
 -------------
 
 Next we get and build DyNet:
 
-    git clone https://github.com/clab/dynet.git /path/to/dynet
-    cd /path/to/dynet
-    mkdir build
-    cd build
+    $ git clone https://github.com/clab/dynet.git /path/to/dynet
+    $ cd /path/to/dynet
+    $ mkdir build
+    $ cd build
 
 Case of using CPUs to calculate neural network:
 
-    cmake .. -DEIGEN3_INCLUDE_DIR=/path/to/eigen
-    make -j <threads>
+    $ cmake .. -DEIGEN3_INCLUDE_DIR=/path/to/eigen
+    $ make -j <threads>
 
 Case of using CUDA:
 
-    cmake .. -DEIGEN3_INCLUDE_DIR=/path/to/eigen -DBACKEND=cuda
-    make -j <threads>
+    $ cmake .. -DEIGEN3_INCLUDE_DIR=/path/to/eigen -DBACKEND=cuda
+    $ make -j <threads>
 
 
-Configuring library paths
--------------------------
+Install MTEval
+--------------
 
-All shared libraries of CUDA and DyNet should be visible from the NMTKit
-binaries.
-Add a configuration in your shell-rc file like:
+MTEval can be installed by similar way to DyNet:
 
-    export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/path/to/cuda/lib64
-    export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/path/to/dynet/build/dynet
+    $ git clone https://github.com/odashi/mteval.git /path/to/mteval
+    $ cd /path/to/mteval
+    $ mkdir build
+    $ cd build
+    $ cmake ..
+    $ make -j <threads>
 
 
 Install NMTKit
 --------------
 
-    git clone https://github.com/odashi/nmtkit.git /path/to/nmtkit
-    cd /path/to/nmtkit
-    git submodule init
-    git submodule update
-    autoreconf -i
+Prepare build tree for NMTKit:
 
-Case of using CPUs to calculate neural network:
+    $ git clone https://github.com/odashi/nmtkit.git /path/to/nmtkit
+    $ cd /path/to/nmtkit
+    $ git submodule init
+    $ git submodule update
+    $ mkdir build
+    $ cd build
 
-    ./configure --with-eigen=/path/to/eigen --with-dynet=/path/to/dynet
-    make
+And configure makefiles with all library locations:
 
-Case of using CUDA:
+    $ cmake .. \
+        -DUSE_GPU=ON \              # need if you use CUDA
+        -DCUDA_ROOT=/path/to/cuda \ # ditto
+        -DEIGEN3_INCLUDE_DIR=/path/to/eigen \
+        -DDYNET_INCLUDE_DIR=/path/to/dynet \
+        -DDYNET_LIBRARY_DIR=/path/to/dynet/build/dynet \
+        -DMTEVAL_INCLUDE_DIR=/path/to/mteval \
+        -DMTEVAL_LIBRARY_DIR=/path/to/mteval/build/mteval \
+
+And then:
 
-    ./configure --with-eigen=/path/to/eigen --with-dynet=/path/to/dynet --with-cuda=/path/to/cuda
-    make
+    $ make -j <threads>
+
+
+Configuring library paths (optional)
+------------------------------------
+
+All linked libraries should be visible from the NMTKit frontend binaries.
+To give library locations explicitly, for example, add configurations in your
+shell-rc file like:
+
+    export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/path/to/cuda/lib64
+    export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/path/to/boost/lib
+    export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/path/to/dynet/build/dynet
+    export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/path/to/mteval/build/mteval
+    export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/path/to/nmtkit/build/nmtkit
 
 
 Validation
 ----------
 
-    make check
+    $ make test
 
 Sample files could be used to validate the behavior of binaries:
 
-    src/bin/train --config sample_data/tiny_config.ini --model model
-    src/bin/decode --model model < sample_data/tiny.in
+    $ /path/to/nmtkit/build/bin/train \
+        --config /path/to/nmtkit/sample_data/tiny_config.ini \
+        --model model
+    $ /path/to/nmtkit/build/bin/decode \
+        --model model \
+        < /path/to/nmtkit/sample_data/tiny.in

doc/training.md

@@ -63,8 +63,8 @@ the location of the output directory.
 
 To launch the trainer with the sample corpus, type following lines:
 
-    src/bin/train \
-        --config sample_data/sample_config.ini \
+    $ /path/to/train \
+        --config /path/to/nmtkit/sample_data/sample_config.ini \
         --model model
 
 `train` command modifies only files in the output directory specified by the
@@ -92,7 +92,7 @@ The prefix of the `*.params` files means:
 We also could output similar contents as `training.log` to stderr by specifying
 `--log-to-stderr` option:
 
-    src/bin/train \
-        --config sample_data/sample_config.ini \
+    $ /path/to/train \
+        --config /path/to/nmtkit/sample_data/sample_config.ini \
         --model model \
         --log-to-stderr