osarenomaspecial:
I hope after everything we build a real world project.

Well, let's see how it goes.
Additionally, different projects can be proposed by different persons (and these projects can be worked on by groups of persons in the community). Therefore, with such system, I hope everyone will have a part to play.

Anybody can put up an idea for a project on the group, with the hope that discussions are organically grown around the project and then these discussions may lead to actual implementation via collaboration by interested members of the group. From this perspective, it means that it's possible for the group to have different projects at an instance in time.

Again, thanks for all comments so far.
The survey ends tomorrow (Sunday) by 11:59PM.

We currently have about 15 persons who have indicated their interest.

Chukwudaalu:
Count me in

Thanks

tdayof:
Can you put this into work on robotics and facial recognition system?

Yes you can.

Again, thanks for the interests shown so far.

The survey continues till December 10th 2017.
Thanks.

Thanks for all the response.

We now have more than 10 interested persons. This is good.

I'll keep the survey running (as earlier stated) till 10th December.
Hopefully, more people will indicate their interest.

Thank you all.

Thanks for all the comments.
So far, we have gotten about 7 interested persons (with mixture of beginners, enthusiasts and people with experience). That's good.

Like I mentioned earlier, I'll try to keep this thread on the front page of programming section for the duration of this survey (10 days).
Thanks.

TL;DR
Will you be interested in a Nairaland thread focused on discussions around Artificial Intelligence/Machine Learning techniques, with some focus on applications in Nigeria? If yes, please drop a comment below. If we get up to 10 interested persons on or before 10th December, I'll be willing to create the group's thread here on Nairaland.
Thank you.


Hi,
This thread is a survey for people with interest in AI/ML.

What we have to far / Current status
There have been some direct/indirect attempts to create/foster a community of people interested in Artificial Intelligence (Machine Learning, Computer Vision, Language Processing and so on) as seen in [1] and [2]. However, nothing really concrete materialized. It will be nice to have an AI community with different levels/mix of knowledge and experience (including people with little or no experience), were discussions ranging from research papers to Nigerian focused AI related projects. I will be interested in such a community.

The Survey / My Proposal
Therefore, I'm taking a survey here on Nairaland to know if people will be interested in such a community. The survey will run for 10 days (from 1st December - 10th December 2017). If we can get up to 10 interested persons (with different mix of experience/knowledge), I'll be willing to create a Nairaland thread (which will serve as a group) were such a community can start, grow and possibly thrive.

and then, the Requirements
To ensure you know what you are getting into, below is a list of requirements for interested persons
a. Interested in discussions/developments/research about modern AI systems only (see note below for description about what counts as morder AI).
b. Serious minded
c. That's all about requirement. :-)

Finally, the reason for this thread
If you are interested in such a community, please drop a comment below.

Important Points to Note
a. I'm not here to teach you about AI/ML. No, it's not a one sided thing. We all contribute and learn from each other about the current AI/ML projects/research that you're are working on. The aim is to foster a community of AI developers/researchers in Nigeria.

b. Modern AI (as opposed to classical AI which includes subjects like travelling salesman problem, breadth first search, simple puzzle solver etc.) focus on Machine/Deep Learning techniques (e.g. neural networks) being applied to areas such as computer vision tasks (e.g. image classification/localization), natural language processing tasks (e.g. language translation), speech recognition tasks (e.g. speech-to-text), control/planning tasks etc.

c. I know there exist AI/ML related forums online that bring together people from different countries (such as this reddit group [3]). However, I feel we need to do something more local to foster an AI community here in Nigeria.

d. I'll try to keep this thread in the front page of the programming section daily by so as to get enough viewers as possible.

Thank you.

Links
[1] https://www.nairaland.com/3371116/shall-discuss-artificial-neural-networks
[2] https://www.nairaland.com/4177343/revolutionizing-world-artificial-intelligence
[3] https://www.reddit.com/r/MachineLearning/

bearnana:
Good gracious... I didn't expect this much focus...
@SoftEng thanks for your input

Just to clarify what I did with my TicTacToe
1: I built a system that takes the moves a user makes and inserts into the database
2: If you win the computer, those moves are regarded as success moves
3: If you play again with computer and computer is to start, it plays that success moves to try to win you.
4: If you still win it repeats step two until the all possible moves has been learnt.

Thats pretty much AI in PHP and MySql

If I wanted to do something like Language Processing, yes it would consider hardware. But think of something like [url="https://www.program-o.com/"]this[/url], it a chatbot. It was built in PHP and MySql. It still follows the same line of logic for the game.



Thanks for the advice, I wonder why I didn't think of JavaScript?

@osarenomaspecial Thanks for the advice.

So far the AI thing is not as confusing as I thought. There is no need to discourage anybody with hardware semantics (my system is an intel dual core and it is serving me well so far). But like I said, my reason for learning AI is because of Quantum Computation (thats where I see some real headache.

OK.
Good luck with your AI and Quantum Computing learning. Also, keep sharing your knowledge on these subjects just as you have already done so far (via your tic-tac-toe game).

osarenomaspecial:
those research lab have access to free GPU for study purposes, like google for example do give free credit to train data. what am trying to say is this, use small data set to study and understand the algorithms and later move on to large data set.

OK.

TheOtherGeek:


Yo, it seems you're mixing things up... Earlier, you said she should use a language that makes it easy to interact with the GPU (Nvidia's CUDA), so someone mentioned using cloud solutions for training data which automatically targets the GPU but you still complained about it most likely being GPU based... (I'm confused, need clarification)

The thing is, although PHP can't target CUDA directly because libcudart.so is not available by default for the web server (or so) read this, So using PHP may be a bad idea even from the start but that doesn't mean you can't work with GPUs because you can't interact with CUDA... There is OpenCL (an alternative) and also a PHP library for it, although it was done in 2012, which is enough reason to back off except you have the mental capability to refactor the codebase, in case it happens to be deprecated (which could be overkill because it'll be most likely deprecated)

But, I'd leave @bearnana with note;
Using OpenCL over CUDA cos of compatibility issues means you have to work 5x harder (if you're building from scratch) so, just give it a thought, you could even use JavaScript if you're more comfortable with it over Python.
Read this when you've got the time;
https://news.ycombinator.com/item?id=12309419

Just don't bother yourself about what people say about large datasets...who cares, you're just learning right? Not a NASA project yet so keep learning and when you notice hops or issues then you can learn how to work with GPUs instead of the traditional CPU.

Moving Away from the topic a bit, who has used UE4?? You prototype your app using VR, that's a crazy feature built in on a C++ engine.. That's top notch machine learning and computer vision , the fact that frameworks such as TensorFlow can be integrated into the engine is cool too, so one could build nice VR experiences using machine learning.

Sorry for the mix up. To clarify, I meant that your subscribed cloud solution likely contain a GPU. Therefore, you will need to use a language that interfaces well with the GPUs. This means that subscribing for a cloud solution with GPU and then using a languagge that does not interface well enough with the GPU is a waste of resources.


We should not be focusing on this issue for too long. Please use any language you want to use. I have not been against this from the beginning for small dataset. Here are some of my comments earlier.

...
On the long run, when you delve deeper and deeper into technical aspects of modern AI (which is all about applying Machine Learning/Deep Learning/Reinforcement Learning techniques to computer vision, natural language processing, speech, robotics and so on), then you may have to use python or java (for enterprise related AI).
...


Lastly, I'll note here that if you are building just a small AI system, then you may not need to consider your hardware...

osarenomaspecial:
you don't need large computational power when learning with small data set,you only need computational power when training large data set for production, it is advisable to use those popular machine learning libraries that has great abilities like scalability and so on.

OK.

But even in research lab these days, you don't deal with small datasets. Take for instance ImageNet dataset for computer vision. Even CIFAR-10 dataset (which is way smaller than ImageNet) can require a decent compute power as your neural network architecture size increases.

4kings:

Good point, but this depends on the datasets. Beginners would normally start small.
But for computational power, there are always cloud services like, AWS(Amazon Web Service), Google cloud platform, IBM Blue mix, Azure and so on.
I use pythonanywhere.

Though i'm not sure if these options will work for all languages(especially php) but you get the point.

OK.

But note that the compute power (to train your AI) that you are getting from cloud providers like AWS is still likely to be GPU-based.

4kings:

Ya, if one knows the maths then choice of language is no problem.

In terms of programming modern AI systems, it's not just about knowing the math. You can also write the low level maths in python. But there's something else you have to seriously consider. Which is your hardware. Mordern AI systems are computationally expensive.

To seriously train modern AIs (basically neural networks), you need to use a GPU (Nvidia GPUs are the most favourable to use because of their underlying very low level libraries called CUDA).

Mordern AI libraries in python are written to interface with some low level C/C++ codes which in turns interface with CUDA library for NVIDIA. So that means that interfacing your AI systems with GPU is taken care of by default for you. That means that you can then write the math equations in python (but they are processed in the GPUs). Not many languages have this kind of software library stack (interfacing with CUDA) like python. I believe Java has this capability thanks to the some AI libraries.

If your AI codes can't run on GPU (let's say because your choiced language don't yet have bindings for GPU support), it means that it will run on CPU. The difference between training your AI system on CPU and GPU is HUGE. For example, a neural network that will take 3 hours to train completely on a GPU can take 18 hours or more to train on CPU.

Lastly, I'll note here that if you are building just a small AI system, then you may not need to consider your hardware. But I'll tell you that modern AI systems that powers Language Processing systems, Computer Vision, Speech etc are nothing near small in computational size.

Fin.

bearnana:


I know that Python is the best choice for AI. I do intend to jump to python when it seems as though PHP may not serve me well. My choice of PHP is because of my current work. I believe before I get to the part where it seems too horrendous for PHP to tackle, I'll quickly pivot to Python.
And there are Machine Learning Libraries in PHP...

OK.

bearnana:
I was reading up on Satoshi's Bitcoin Algorithm and I stumbled on Quantum Computing which led me to AI. So I have decided to tow the line of AI. My aim is to contribute to Quantum Computing. So I created a github repo which I will try to contribute to everyday until I feel I have grasped AI and ready for Quantum Computing.

I know there are some devs that want to jump into AI but are a little confused so if you want some guidance from a newbie, I'd be updating the readme everyday before I go to bed. I hope there is another psycho out there like me who is also adventurous.

FYI, I started officially yesterday. I wrote a Tic Tac Toe game in PHP and codeigniter. So if you want to catch up, that was my task yesterday and I'm done today, so I'll drop what I did today tomorrow.

Github link : https://github.com/Bear-nana/AI-Ignorance

I looked through your github page.
AI is a good field to get into. I have been reading about some promising news about Quantum Computing as well, so I guess something practical will soon be coming out of the field that has been in theory mode for a long time.
On the long run, when you delve deeper and deeper into technical aspects of modern AI (which is all about applying Machine Learning/Deep Learning/Reinforcement Learning techniques to computer vision, natural language processing, speech, robotics and so on), then you may have to use python or java (for enterprise related AI). The main reason to use python is because most of the really good libraries for modern AI are available in python. Although I have not checked, I suspect that you won't get any good AI library in PHP (except for the ones that may help you call APIs that contain pre-built AI models).

This is good.
AI is going to solve interesting problems in the world. The idea of a research lab is not bad. However, I will be more interested in a research lab that also has a research application deployment unit that applies research to solve practical societal challenges.

Let us also note that there are some problems that AI cannot fix for us (e.g. current electricity challenge and poor roads are not direct problems AI will solve)

P.S.
A friend and myself started a blog to write about AI and it's benefits to Nigeria. We aim to help stimulate and build AI interest in Nigeria, which can in turn help foster a community. I will share the link soon.

Additionally, another effort of AI community building worth mentioning is by the guys from Forloop (http://forloop.africa/). The recently started a ForloopAI movement (https://twitter.com/forloopai), which I believe will host meetups/conference in different cities to start a community of AI development. I believe they will deliver on this going by what they have done with general developer community building.

osarenomaspecial:
is there community in that Area in Nigeria.

I believe there is a community (although still very young). It's called Data Science Nigeria. Here's a link to their webiste and twitter handle. OneFi is one of the key organizations spearheading this movement and they had a loan risk prediction challenge on kaggle (see link below)

website - http://www.datasciencenigeria.org/
twitter - https://twitter.com/DataScienceNIG
kaggle challenge - https://www.kaggle.com/c/data-science-nigeria-credit-risk-prediction

It is also possible that there other Data Science communities existing in Nigeria that I haven't heard about.

4kings:
I have two algorithm requests.

1) Thresholding for multiple color detection.

2) Algorithm to extract location of edges after performing edge detection on images.

(Preferably without OpenCV just PIL and maybe numpy. But hey beggars are not choosers )

Cc: Emrain , UmaryUsuf , Seun Adepeter2027 , Deicide , sonofluc1fer , JonBellion osarenomaspecial , SoftEng , adejumoadeoluwa , melodyogonna classicdude Olyboy16 pcguru danielthegeek guru01 and others.

---

Non-pythonitas can also post their codes so long as it's explained.

I hardly do work on image processing. However, here's what I was able to find for #2

from PIL import Image, ImageFilter
import numpy as np

im = Image.open('your-image.jpg')

# Edge detection: background should be black (i.e. pi    xel value = 0). only edges should be set to a non-zero pixel value
im = im.filter(ImageFilter.FIND_EDGES) 
im.save('output.jpg')

# Therefore to get the indexes of edges, retrieve the indexes of non-zero pixels.
np_im = np.array(im)
indexes = np.where(np_im != 0)  
print(indexes)

Note: that you can also use ideas from the scipy link specified by HottestFire (in fact this should be preferable, as the canny edge detection algorithm in scipy *might be* better than PIL's edge detection implementation). All you need to do at the end is to get the indexes that are non-zero as shown above.

osarenomaspecial:
thanks. are you a Researcher or DL Developer.

Kind of both.
Applying DL to practical challenges (as a DL developer) is nice, and research also is interesting.

At the moment, anyone that wants to be a (low level) DL developer still have to do research (or at least keep an eye on current research) due to the current nature of the field. This will likely change in the near future, especially for applications relating to supervised learning.

aakpan904 is right about derailing this thread.

I won't make any more DL related comments on this thread.
I guess starting a DL thread is worth considering... (I guess it should depend on the number of people interested in it, to start a micro DL community)

Thanks.

osarenomaspecial:
am using tensorflow, pytorch and Python numpy. only use numpy when you are good with the theories and maths but those libraries do that for you with ease. your guess are right, as I earlier said it a very complex work, up till now am still looking for ways to solve it. I downloaded a paper yesterday which I believe will give me clue to solve the problem, the paper is on reinforcement learning. that is the reason I want to learn reinforcement learning.

OK.
I've only recently tried implementing simple stuffs in raw numpy and then theano (following through a blog post).
Primarily I use [url]keras.io[/url](and then I switch been tensor flow and theano as backend for keras)

With regards to your RL study, I'll advice you read some introductory textbook or blog posts before diving into papers (I guess you've already done this). Also, another good strategy is to go through papers that write about overview of RL.

Additionally, I'll put up some links here (You may already know some of them) for anyone interested in Reinforcement Learning.

1. Richard Sutton's (draft copy) book is a good introduction -
http://incompleteideas.net/sutton/book/bookdraft2016sep.pdf

2. For those that prefer video tutorial - David Silver's intor to RL lecture is on youtube.
https://www.youtube.com/watch?v=2pWv7GOvuf0&list=PLzuuYNsE1EZAXYR4FJ75jcJseBmo4KQ9-

(From what I read online, the video lecture mostly followed the content of Richard Sutton's book)

3. Lastly recent RL overview paper on arxiv
https://arxiv.org/abs/1708.05866
https://arxiv.org/abs/1701.07274


Lastly, I apologise to OP for derailing this thread which was a question about python2/3 syntax.
It was nice to see people enthusiastic about DL and I had to jump in.

michelle5683:


It's seems as though artificial intelligence is a large and unexploited part of python

Yeah, artificial intelligence (A.I.) is getting quite large large these days.
I'll like to add that you can use other languages as well for A.I., but python (libraries that has low level C/C++ bindings) seems to be by far the most popular in recent times.

Lastly, the cool thing about python programming is that you can learn the basics (as you are doing right now), and branch off to different areas of programming of your choice. (e.g. A.I., web programming, network programming, game programming - like your guessing game, systems programming, dev ops)

osarenomaspecial:
i have worked on Image recognition using deep learning but now am currently working on crime detecting and tracking system using artificial intelligence. to train the system to recognize crime but I discovered it was very complex, it involved machine learning, deep learning and reinforcements learning am not good with reinforcement learning, that is what I want to learn now.

Had to login because of this...
Nice to see that people are getting into deep learning (DL).
I have also been reading a lot about it for about a some time now (with some small image classification experiments on cifar-10 image dataset).

I have 2 questions for you.
1. Since you use python, do you use a DL framework/library. If yes, which one?
2. How does reinforcement learning (RL) fit into your project for crime dectection?

My guess for #2 is that you may want a system that can also 'act' and not just classify scenes as 'crime' or 'not crime'. If this is the case, then maybe I guess developing the system to 'act' by calling the appropriate security authority might be one of the angles you're looking at it from.

Anyway, I'm just guessing. I await you reply.

Thanks.


Lastly, @OP
Keep up the good work.
We need more people to take up interest in software (and also hardware) in this country.
Even more important than the above statement, we need more people to start using their skills to solve practical problems in our society.

Well done.

TL; DR
Removing the zeros trailing each number in your else if condition


The Issue
The issue is with the way you represent the numbers (i.e. 01, 02, 03 and so on) in the if condition.
Integer number are usually not represented that way in code.
It was strange to see that php allowed it.

After some little experiment in PHP REPL and a little online search, I realised that I you prefix an integer with zero in PHP, it menas that the number is an 'octal' (base number. This means number like 08 and 09 is not possible to represent because they are not part of the octal number system. Therefore, write 08 or 09 in your code is an error.

From the stack overflow link below, it looks like error such as the one described above was ignored in PHP 5, however, in PHP 7, codes with such errors produces an error message
https://stackoverflow.com/questions/40735963/parse-error-invalid-numeric-literal


The Solution
In essence, to resolve the bug in your code, remove the trailing zero before each number (in the if-condition) in your code. Your code should be:

...

else if ($seg == 1 || $seg == 2 || $seg == 3 || $seg == 4 || $seg == 5 || $seg == 6 || $seg == 7 || $seg == 8 || $seg == 9) {

...

}

...