Forem: dinhanhx

Understand n dimension array with real life examples

dinhanhx — Sun, 27 Feb 2022 12:51:38 +0000

Scenario

You are/were struggling with one of these terms: nested array, high dimension, n dimension array, ... Probably you have tried to visualize an n dimension array and got stuck at 3 dimension or even 2.

The simple way to understand

You don't need to visualize it. You just need to know that an n dimension array is a group of groups ... (many "of groups").

Let start with 1D array. 1D array is a group of values [1, 2, 3]. 2D array is a group of 1D groups [[1, 2, 3], [1, 2, 3]]. 3D array is a group of 2D groups [[[1, 2, 3], [1, 2, 3]], [[1, 2, 3], [1, 2, 3]]].

An n dimension array is also just a 1D array whose values are organized into groups. This is 1D array [1, 2, 3, 4, 5, 6]. Now we will group 123, and 456. By doing that, it becomes 2D array [[1, 2, 3], [4, 5, 6]]. Now we put each value into a group. As a result, 3D array is made [[[1], [2], [3]], [[4], [5], [6]]].

Every array has a very important attribute called shape. A shape is a set of the number of groups/values in each dimension (ordered from the highest dimension to the lowest one). Let visit the previous example. 1D array has 6 values or shape (6). 2D array has 2 groups, each group has 3 values. In other word, its shape is (2, 3). That 3D array has shape (2, 3, 1).

Real life example 1: A book

A book is a 5D array of letters whose shape is (c, p, s, w, l). A book has C chapters. Each chapter has P paragraphs. Each paragraph has S sentences. Each sentence has W words. Each word has L letters.

Real life example 2: A library

A library is a 7D array of letters. A library has G genres. Each genre has B books. And so on.

How to X paper

dinhanhx — Wed, 26 Jan 2022 11:05:07 +0000

Dear machine learning newbies, please don't skip steps

dinhanhx — Mon, 24 Jan 2022 11:15:44 +0000

I know machine learning is a hot topic. You probably are motivated by news, social media saying how smart an AI system is. Your parents might have advised you to follow this field. Your smarty pant friend writes a very smart program to analyse weather. Maybe you, come from other majors than computer science, are influenced by your own professors.

Then you decide to google "how to study machine learning" or "good university for AI major" or "how to make an assistant like Jarvis" despite the fact that you don't even know how to write a simple program. You probably don't understand the following abbreviations: CLI, OS, OOP. Magically, you copy code just like a monkey. It works, but you don't understand WHY.

Basically, this is you -> https://youtu.be/MNb9Q8vAzjw?t=49

Congratulations, you have taken shortcuts. Everyone is happy and clapping. And that's my sarcasm.

If you are a self-study machine learning newbie, please read the requirements of the course, learn the requirements, then learn the course.

For example, you take Machine Learning by Tutorialspoint.

See that prerequisites, learn all that first.

If English is not your first language, learn English vocabulary for Computer Science. Most of the documents about Machine Learning are in English. Don't trust me? Just ask your any fellow people who have been working or learning things related to computer.

If you are following this major by going to university, focus on class, close this blog, start doing your assignments.

How stupid is Matlab?

dinhanhx — Fri, 14 Jan 2022 03:49:41 +0000

Matlab Introduction from MathWork

It's for Math, but

Array indexing starts from 1. Ask your professors from Electrical Engineering which number signal indexing starts at? They will probably open textbook to show or to say it's 0. Or ask your Applied Math professors where is the axis's origin at?
Functions are defined after being called in script. I'm pretty sure that we define/state a formula first, then we use it when doing math on paper.

It's for Graph, yeah.

It's for Programming, but

More proofs of how stupid is Matlab:

Matlab vs Python comparison page.

The following picture is captured from MathWork's page about comparing to Python.

Here you see they wrote 2 sentences describing how awesome their products are. And they just wrote 1 sentence describing Python. Then they went on how it's suitable for engineering and scientits from other fields other than computer science.

I'm pretty sure that is not how you should compare things, MathWork. Let take an example Matlab vs Python by RealPython in which people fully compare things feature by feature.

Matlab pricing

It's stupidly expensive if you are a scientist in a country that isn't rich as top economic countries.

How stupid is Matlab function?

dinhanhx — Fri, 14 Jan 2022 02:52:24 +0000

Function definition in a script

Python

Filename: the_amazing_file.py

def returnTrue():
    return True

print(returnTrue())

Here you define the function first, then call it. It's like how you do math with paper. You write down the formula, then you apply it.

Matlab

So the language for math, the function should be defined first right?

Yeah no.

Filename: the_amazing_file.m

disp(returnTrue());

function y = returnTrue()
    y = true;
end

Seriously matlab, are you for math?

Calling non-parameter function

Let say you have a function foo() which doesn't have parameters

bar = foo() 
bar = foo
% Both are equivalent.

But foo() and foo mean two different things. foo() means calling a function. foo means "I'm a variable". If you use Python, it's passing functions as an argument. If you use C, it's function pointer.

Passing functions as argument vs Defining anonymous functions

Alright how to pass function as an argument in matlab? You use @ I copied this example from matlab function doc

% Compute the value of the integrand at 2*pi/3.
x = 2*pi/3;
y = myIntegrand(x)

% Compute the area under the curve from 0 to pi.
xmin = 0;
xmax = pi;
f = @myIntegrand;
a = integral(f,xmin,xmax)

function y = myIntegrand(x)
    y = sin(x).^3;
end

Guess what else use @? Anonymous functions

sqr = @(x) x.^2;

If you look at the way anonymous function is defined and look at how non-parameter function (previous section), you may ask the following line

bar = foo;

could mean anything right? Yeah it could mean:

calling a function without parameters
assigning anonymous function to different variable.

The more confusing, the more we analyse.

Now you may argue, "I know my functions". Yeah but can you remember other people functions when reading a long file while trying to understand what it does? Can you?

How stupidly expensive is Matlab for Deep Learning?

dinhanhx — Wed, 12 Jan 2022 06:31:36 +0000

Thank to Xuan Vu Nguyen for some discussions.

With Python, you have Tensorlow 2/Keras (free), PyTorch/Lightning (free), Jupyter Notebook (free). You can even use Google Colab if you don't own a decent gpu.

With Matlab (perpetual license), you have to buy Matlab for 2.3K USD

DeepLearning ToolBox for 1.3K USD

and to use that tool box with gpu, you have to buy Parallel Computing Toolbox for 1.1K USD

Not mention that Matlab often consume much hardware resources so you may need a better cpu, ram, hard disk. (You guess right, you have to pay even more).

MathWork, do you even research what people demand? What people need? No one is going to pay 4.7K USD for a deep learning enviroment with matlab (a crappy framework with a stupid language) because they spend money on GPU already.

How stupid is Matlab OOP?

dinhanhx — Wed, 12 Jan 2022 05:57:15 +0000

Let write a normal class in Java first

public class Car {
    private int wheels;
    public void setWheels(int new_wheels) {
        wheels = new_wheels;
    }
}

Keep in mind that Java is a lower-language than Matlab. Now you might ask in Matlab it should simpler than the Java code. Guess what. It's no.

classdef Car
    properties (Access = public)
        wheels
    end
    methods (Access = private)
        function setWheels(new_wheels)
            wheels = new_wheels
        end
    end
end

In Matlab, you have to define blocks (methods and properties) to separate functions and variables. What even more confusing is that MathWork calls Access an attribute. (Method Attributes, Properties attributes). In UML notation or any language providing OOP, attributes mean object/class variables or properties. The right words for public or private are modifiers (in Java), specifiers (in C++). As a result, when you are talking with other engineers (who are not proficient in matlab), you could make them confused with words.

Now we will add a public integer attribute plate to class Car

in Java

private int plate;

in Matlab

properties (Access = private)
    plate
end

You see unnecessary keywords and characters in matlab. If the researchers or the engineering need fast prototyping, matlab isn't for them. Ironically, Matlab claim to be the right language for scientific computing.

How stupid is Matlab accessing array syntax?

dinhanhx — Wed, 12 Jan 2022 04:15:50 +0000

Here is how you define an array and print the first element in it

v = [4 5 6 7];
disp(v(1));

Let not talk about the fact that matlab array starts with 1 for now.

So what's the problem here? The functions also use () for receiving parameters. When you read this line without previous lines, disp(v(1)) you have to ask yourself this question:

v is a function or an array?

"I always read the previous lines" you may argue. But can you remember which one is a function, which one is an array in a long script with many functions and variables?

Now you have to remember () has two meanings. Well two usage of () near each other can create confusion.

How stupid is Matlab for loop?

dinhanhx — Tue, 11 Jan 2022 12:39:55 +0000

Before I answer the question in the title, let look at how for loop in Python and C are expressed in human language

Python

v = [1, 2, 3, 4]
for i in v:
    print(i)

Human language: there is a variable v assigned a list [1, 2, 3, 4]. Then for each element in v (stored in variable i) is printed.

You can see it's very straight forward.

C

int v[] = {1, 2, 3, 4};
int i = 0;
for(i = 0; i < 4; i++) {
    print("%d", v[i]);
}
//for ( init; condition; increment ) {
//   statement(s);
//}

Human language: there is a variable v assigned an array {1, 2, 3, 4}. Then, a variable i stores integer 0. Then for i starts with 0 (the initial), when is smaller than 4 (the condition), increase i by one (the increment) and print the value element of v by the index (the statment).

It's longer than Python (of course, C is a low level language). However, the description is still straightforward. It's just added some linking words to ease human mind.

Matlab

for v = [1 2 3 4]
   disp(v);
end

By instinct, when you see v = [1 2 3 4], you will say a variable v assigned an array [1 2 3 4]. Guess what? it becomes the iterator when being added to a for loop.

I tried to express this thing in human language that is close to the code but nothing works.

Try 1: for v is assigned to an array, display element of v

Here you can see it doesn't show that v is an iterator. And v has two meanings.

Try 2: for each v in [1 2 3 4], display v

But v is assigned to an array?? We all know equal sign and belong sign are two different things.

There isn't a proper human language description for this one. And you have to remember that = has two meanings:

assign when not in a for loop
in when in a for loop

If MathWork read this blog, please use something different than =. You can use in or belongto or iterate.

Now you see how stupid it is.

Attention this, attention that

dinhanhx — Wed, 16 Jun 2021 03:07:29 +0000

This post is for people who are working on or learning in deep learning with natural language processing. Minimum knowledge level: Hugging Face - Transformer or equivalent.

Who have read 3 followings papers are beneficial:

There are 3 style of attention mechanism should not be confused with namely:

Name	Tensorflow implementation	PyTorch implementation	Paper
Bahdanau attention	AdditiveAttention layer	Attention Decoder	Neural Machine Translation by Jointly Learning to Align and Translate
Luong attention	Attention layer	Decoder's Attention	Effective Approaches to Attention-based Neural Machine Translation
MultiHeadAttention	MultiHeadAttention layer	MultiheadAttention layer	Attention Is All You Need

Reading list recommendation:

NOTE: Please comment any attention mechanism not included in this post as well as paper, implementations.

My workflow for a research in machine learning

dinhanhx — Tue, 15 Jun 2021 08:52:04 +0000

This post is based on my experience when I was an intern at USTH ICT Lab. It's aimed for everyone to read.

I divide the research into 3 stages.

Stage 1: Preparation

Define in detail what the problem is.
Define performance metrics.
Find dataset and relevant datasets.
Read top research papers related to the problem as much as you can but not too deep. Read abstract, read model overview, read results.
Check their source code if there are.
Summarize all the ideas, all results and all hardware specification.

Stage 2: Experiments

Pick 2 or 3 research papers that you feel you can reproduce the results.
Read them in detail.
Reproduce results.
Brainstorm idea to what can be done to improve the result.
Carry out your experiments.
Summarize your work.

Stage 3: Report

Write a research paper
Publish your work.

The way you should install C++ libraries and install Visual Studio on Windows 10

dinhanhx — Thu, 25 Mar 2021 08:56:02 +0000

In this tutorial, I will show you the way to install Visual Studio as small as minimum as possible.

Install Visual Studio

Download Visual Studio Installer

When in Visual Studio Installer GUI windows, I suggest you tick like this

It will consume your diskspace at most 4gb.

Install vcpkg

vcpkg is a cross-platform C++ libraries manager. It will do the hard work for you such as building, importing, setting up PATHS, ..etc.

How to install vcpkg

To install vcpkg

> git clone https://github.com/microsoft/vcpkg
> .\vcpkg\bootstrap-vcpkg.bat

To install the libraries for your project, run:

> .\vcpkg\vcpkg install [packages to install]

You can also search for the libraries you need with the search subcommand:

> .\vcpkg\vcpkg search [search term]

In order to use vcpkg with Visual Studio,
run the following command (may require administrator elevation):

> .\vcpkg\vcpkg integrate install

All installed libraries are immediately ready to be #include'd and used in your project without additional configuration.

Example

I want to setup glut, glew, glm for my Computer Graphics course.

I go to where my vcpkg is installed. Then I run

vcpkg.exe install freeglut glew glm

I wait until it is done.

vcpkg.exe integrate install