N-Gram ARPA Model

This post is to summarize how the probability is calculated from ARPA model.

Consider test.arpa and the following sequences of words:
look beyond
more looking on

Let's consider look beyond first. The log10 probability of seeing beyond conditioned upon look, i.e., log10(P(beyond | look)) = -0.2922095. This is directly from the test.arpa file, line 78.

What is, then, the probability of seeing look beyond? Well, this is by the chain rule of conditional probabilities

log10(P(look beyond))
= log10(P(look) * P(beyond | look)) 
= log10(P(look)) + log10(P(beyond | look)) 
= -1.687872 + -0.2922095 = -1.980081558227539, 

which can be verified with python code

import kenlm
model = kenlm.LanguageModel('test.arpa')
print(model.score('look beyond', eos=False, bos=False)


Let's try the next sequence more looking on. Let us start with the chain rule

log10(P(more looking on))
= log10(P(more)) + log10(P(looking | more)) + log10(P(on | more looking))

The first term on the RHS is easy: log10(P(more)) = -1.206319 from line 34

The second term is a bit tricky, because we cannot find the bi-gram more looking from the model. Hence, we use the following formula:
P(looking | more) = P(looking) * BW(more)
where log10(P(looking)) = -1.285941 from line 33, and log10(BW(more)) = -0.544068 is the back-off weight, which can be read off from line 34.

Lastly, the third term is again not present in the model, so we reduce it to
P(on | more looking) = P(on | looking) * BW(looking | more)
where the first term is -0.4638903 from line 80, and the second term is assumed to be 1, because the bigram more looking does not exist in the model

Thus, we get log10(P(more looking on)) = -(1.206319 + 1.285941 + 0.544068 + 0.4638903) = -3.5

For more details, refer to this document. I also find this answer very helpful.

Remote Debugging with Eclipse

I am not familiar with Eclipse, as I prefer to use CLion. However, for projects that do not use CMake build system, I will have to use Eclipse.

In this post, I will discuss how to remote-debug with Eclipse. The setup is as follows:

target (local): running the application from, say terminal
host (Eclipse): debug the program as it is running

First, open up the project with Eclipse. Make sure that Eclipse can build the project.

Next, setup gdbserver on the target:
$ gdbserver :7777 EXECUTABLE ARG1 ARG2 ...

Here, 7777 is the port we will use for remote-debugging, EXECUTABLE is the binary file we are going to debug as it is running, and ARG1, ARG2, ... are appropriate arguments for this program.

Next, we setup Eclipse debugging.
From the menu, select Run --> Debug Configurations... --> C/C++ Remote Application (double click) --> Using GDB (DSF) Auto Remote Debugging Launcher (Select other) --> GDB (DSF) Manual Remote Debugging Launcher --> OK. Basically, we have selected "manual" remote debugging configuration here.

Make sure Project and C/C++ Application fields are properly filled, i.e., you should be able to select the project from the drop down menu if the project import/build is successful, and choose the EXECUTABLE for C/C++ Application.

In the Debugger tab --> Connection tab, change Port Number to 7777.

Finally, click Debug button. You now should be able to remote debug with Eclipse.

Happy hacking!

Debugging Python Module Built with Pybind11

Pybind11 is a wonderful library to build native c++ functions callable within Python. Let's see how we can debug a module built with Pybind11.

For this project, we will use the very simple Pybind11 example module. Let's get started.

Clone the repository
$ git clone https://github.com/pybind/python_example.git
$ cd python_example

Install the package
$ python setup.py install

Run gdb on Python
$ gdb -q python

Break into the desired function, such as
(gdb) b add
(gdb) b src/main.cpp:4

Run the command
(gdb) r tests/test.py

That is it!

Network Visualization with Intermediate Layer Shapes

Network visualization is very important. Looking at the network graph is much easier than reading through the network description. In this post, I will discuss how to create visual graph of a (Pytorch) network using Netron and include intermediate layer shapes as well.

Netron is an excellent tool for network visualization. At the moment, however, Netron does not support Pytorch natively (experimental feature but not stable). The best thing is to convert Pytorch model to ONNX and then use Netron to graph it.

The following is an example code that graphs ResNet50.
import torch
from torchvision import models
import onnx
from onnx import shape_inference

DEVICE = 'cuda:1'
PATH = 'resnet.onnx'
model = models.resnet50().to(DEVICE)
model.eval()
dummy_input = torch.randn(1,3,224,224).to(DEVICE)
torch.onnx.export(model, dummy_input, PATH, verbose=False)
onnx.save(shape_inference.infer_shapes(onnx.load(PATH)), PATH) # this is required for displaying intermediate shapes



To graph it, simply download Netron and run
$ netron resnet.onnx

That's it!

Resolving Bluetooth Issues with Ubuntu 18.04 LTS

So there are a few things that I do not like about Linux in general; quite a lot of things do not work automatically. One of the problems I have faced is that my Microsoft Designer mouse won't work with Ubuntu 18.04 LTS. After some research, I found this solution suggested by Newbuntie to work well, so I'm going to post this here for any other people going through the same trouble.

Basically, enter the following commands and you are good to go.
$ sudo add-apt-repository ppa:bluetooth/bluez
$ sudo apt install bluez

Enjoy!

Use pkg-config with CMake

Say you want to build a binary which uses OpenCV library. For example, the following would be a typical command for compiling your program
$ g++ `pkg-config --cflags opencv` main.cpp `pkg-config --libs opencv` -o opencv_example

Now, the question is how do we do this with CMake? The following would be what you would write in CMakeLists.txt

project(opencv_example)

# this is where we get pkg-config info

find_package(PkgConfig REQUIRED)

pkg_check_modules(OPENCV REQUIRED opencv)

# this is where you compile your app

add_executable(opencv_example main.cpp)

target_link_libraries(opencv_example ${OPENCV_LIBRARIES})

target_include_directories(opencv_example PUBLIC ${OPENCV_INCLUDE_DIRS})

target_compile_options(opencv_example PUBLIC ${OPENCV_CFLAGS_OTHER})

Happy hacking!

Learn Swift on Ubuntu: install and compile

Swift is a new language developed by Apple. Recently, I am studying Swift so that I can build iOS or macOS apps. In this post, I will go over how to setup Swift environment on Ubuntu 18.04 LTS.

First, install clang
$ sudo apt-get update && sudo apt-get install clang -y

Next, download Swift from Swift.org
$ wget https://swift.org/builds/swift-4.2.1-release/ubuntu1804/swift-4.2.1-RELEASE/swift-4.2.1-RELEASE-ubuntu18.04.tar.gz

Decompress
$ tar xfz swift-4.2.1-RELEASE-ubuntu18.04.tar.gz

Add swift binary directory to your PATH environment
$ export PATH=/path/to/swift/usr/bin:${PATH}

This is it for setting up the environment. Let's now build "hello world" in Swift.

Create hello.swift with the following content:
print("hello world")

To run your code, simply run
$ swift hello.swift
hello world

To compile and create a binary, run
$ swiftc hello.swift
./hello
hello world

That's it for this post. Happy hacking!

Simple 2-Way Chat with Java

In this post, I'm going to discuss how to create a simple chat application with Java. Basically, this is going to be a tutorial on Socket API and Java Thread API.

Here is the basic idea. A chat class should be running two tasks simultaneously: one is to send messages that the user inputs, and the other is to receive incoming message from the other side. Because these two tasks should be run simultaneously, we will need to implement a multi-threaded application.

I am going to let the main thread take care of user input and transmitting over to the other side, and a separate thread for receiving any incoming messages. Here we go.

So, the base class is ChatThread, which pretty much takes care of what I described above. The main thread takes care of user input and transmitting over to the other side, whereas a separate thread is going to run in the background, which receives incoming messages.

For this 2-way simple application, there are only two sides: one for server and the other for client. The server needs to take care of creating the server socket and wait for a client to join. Once the client joins, the server starts the ChatThread. The client will simply join the server and start the chatThread.

The implementation here is very basic and doesn't take care of properly exiting the connections. However, this should give you good introduction as to how to deal with socket programming and threading.

Happy hacking!

Build AI Camera App (Caffe2)

AI Camera app from Facebook is an example app that shows how to build Caffe2 on Android platform. Unfortunately, you will find that it does not work with the latest Android Studio. Here is how to get it to work.

One of the main reasons that it does not build is because starting with NDK r18b, GCC has been removed. The easiest way to get it working is to download older version, say r17c, and edit local.properties file and change ndk.dir to r17c folder.

Other than that, a couple of minor changes. In gradle.build file, update gradle version to 3.2.1 as below:

classpath 'com.android.tools.build:gradle:3.0.1'
classpath 'com.android.tools.build:gradle:3.2.1'

Lastly, edit gradle/wrapper/gradle-wrapper.properties file and change gradle tool version to 4.6 as below:

distributionUrl=https\://services.gradle.org/distributions/gradle-4.1-all.zip
distributionUrl=https\://services.gradle.org/distributions/gradle-4.6-all.zip

After all these changes, the app should successfully build! Happy hacking!

Build Caffe2 using Android NDK

Starting with Google's Android NDK r18b, gcc has been removed (history). That is, only one has to use clang as the toolchain. This may cause some problems if you do not set its toolchain as clang.

In fact, when you try to build Caffe2 for Android with android-ndk, it will fail, complaining

GCC is no longer supported.  See
  https://android.googlesource.com/platform/ndk/+/master/docs/ClangMigration.md.

The solution is to simply edit pytorch/scripts/build_android.sh file, replacing gcc to clang for ANDROID_TOOLCHAIN variable, as shown below:

CMAKE_ARGS+=("-DANDROID_TOOLCHAIN=gcc")
CMAKE_ARGS+=("-DANDROID_TOOLCHAIN=clang")

Happy hacking!