Merge pull request #105 from you-n-g/improve_doc

update documents

binding/lua/README.md

@@ -1,5 +1,10 @@
 # Multiverso Torch/Lua Binding
 
+## Introduction
+Multiverso is a parameter server framework for distributed machine learning.
+This package can leverage multiple machines and GPUs to speed up the torch
+programs.
+
 ## Requirements
 Build multiverso successfully by following the [README > build](../../README.md#build).
 

binding/lua/docs/BENCHMARK.md

@@ -30,13 +30,11 @@ Perform CIFAR-10 classification with torch resnet implementation.
 
 ## Results
 
-| Code Name | #Process | #GPU per Process | Use multiverso | Seconds per epoch | Best Model            |
-| :-------: | :------: | :--------------: | :------------: | :---------------: | :-------------------: |
-| 1P1G0M    | 1        | 1                | 0              | 20.366            | top1:7.288 top5:0.218 |
-| 1P4G0M    | 1        | 4                | 0              | 10.045            | top1:7.704 top5:0.289 |
-| 4P1G1M    | 4        | 1                | 1              |  6.303            | top1:8.610 top5:0.273 |
+| Code Name | #Process(es) | #GPU(s) per Process | Use multiverso | Seconds per epoch | Best Model |
+| :-------: | :----------: | :-----------------: | :------------: | :---------------: | :--------: |
+| 1P1G0M    | 1            | 1                   | 0              | 20.366            | 92.712 %   |
+| 1P4G0M    | 1            | 4                   | 0              | 10.045            | 92.296 %   |
+| 4P1G1M    | 4            | 1                   | 1              |  6.303            | 91.390 %   |
 
 ![top1error_vs_epoch](./imgs/top1error_vs_epoch.png)
-![top5error_vs_epoch](./imgs/top5error_vs_epoch.png)
 ![top1error_vs_runningtime](./imgs/top1error_vs_runningtime.png)
-![top5error_vs_runningtime](./imgs/top5error_vs_runningtime.png)

binding/python/README.md

@@ -1,19 +1,26 @@
-# Installation
-
-## On linux
-Please follow the [README](../../README.md#build) to build and install multiverso.
-
-## On windows
-You need MSBuild.exe installed and make sure your system can find it in the $PATH. Then you should run [build_dll.bat](../../src/build_dll.bat) to build the .dll file and install the .dll. Multiverso doesn't have auto-installer for windows now, you have to copy the .dll to either system $PATH or the multiverso package folder.
+# Multiverso Torch/Lua Binding
 
 
-# Run Unit Tests
+## Introduction
+Multiverso is a parameter server framework for distributed machine learning. This package can leverage multiple machines and GPUs to speed up the python programs.
+
+
+## Installation
+
+1. (For GPU support only) Install CUDA, cuDNN according to this [guide](https://github.com/Microsoft/fb.resnet.torch/blob/multiverso/INSTALL.md). You just need finish the steps before [Install Torch](https://github.com/Microsoft/fb.resnet.torch/blob/multiverso/INSTALL.md#install-torch).
+1. Install the multiverso
+    * On linux: Please follow the [README](../../README.md#build) to build and install multiverso.
+    * On windows: You need MSBuild.exe installed and make sure your system can find it in the $PATH. Then you should run [build_dll.bat](../../src/build_dll.bat) to build the .dll file and install the .dll. There isn't auto-installer for windows now, so you have to copy the .dll to either system $PATH or the multiverso package folder.
+1. Install python binding with the command `sudo python setup.py install`
+
+
+## Run Unit Tests
 ```
 nosetests
 ```
 
 
-# Documentation
-* [Experiments](./docs/EXPERIMENTS.md)
+## Documentation
 * [Tutorial](./docs/TUTORIAL.md)
 * Api documents are written as docstrings in the python source code.
+* [Benchmark](./docs/BENCHMARK.md)

binding/python/docs/BENCHMARK.md

@@ -1,11 +1,11 @@
-# Experiments
-
-## Codebase
-[Deep_Residual_Learning_CIFAR-10](../examples/theano/lasagne/Deep_Residual_Learning_CIFAR-10.py)
+# Multiverso Python Binding Benchmark
 
 ## Task Description
 Perform CIFAR-10 classification with residual networks implementation based on Lasagne.
 
+## Codebase
+[Deep_Residual_Learning_CIFAR-10](../examples/theano/lasagne/Deep_Residual_Learning_CIFAR-10.py)
+
 ## Hardware
 |||
 | -------- |:--------:|
@@ -49,13 +49,12 @@ Clarification
 # The results
 The results of 4 experiments with different configurations are shown as following.
 
-|Short Name | With multiverso | Number of Process(es) | Number of GPU(s) | Sync every X minibatches |  Best model validation accuracy | Time per epoch / s |
+|Short Name | # Process(es) | #GPU(s) per Process | Use multiverso | Sync every X minibatches | Seconds per epoch |  Best model validation accuracy |
 | :---- | :-----: | :-----: | :-----: | :-----: | :-----: | :-----: |
-| 0M-1G | 0 | 1 | 1 | --| 92.61 % | 100.02|
-|1M-1G-1S | 1 | 1 | 1 | 1 | 92.61 % | 109.78|
-|1M-4G-1S | 1 | 4 | 4 | 1 | 92.15 % | 29.38|
-|1M-4G-3S | 1 | 4 | 4 | 3 | 89.61 % | 27.46|
+| 1P1G0M | 1 | 1 | 0 | -- | 100.02 | 92.61 % |
+|1P1G1M1S | 1 | 1 | 1 | 1 | 109.78 | 93.03 % |
+|4P1G1M1S | 4 | 1 | 1 | 1 | 29.38 | 92.15 % |
+|4P1G1M3S | 4 | 1 | 1 | 3 | 27.46 | 89.61 % |
 
 ![accuracy_epoch](./imgs/accuracy_epoch.png)
 ![accuracy_time](./imgs/accuracy_time.png)
-![time_epoch](./imgs/time_epoch.png)