themonk:
Lets go there. I'm higly intrested.
Please as a suggestion, could you mention all intrested persons when you start or can we use a platform like whatsapp. I just need to be duly notified on each update, thats all. Thanks alot

Sure thing. we can do that when the thread for the group is created.

With regards to platform, I believe it should be the decision of the group when created.

Thanks.

4kings:
*coughs*
The day has come.

SoftEng:


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.

INTERESTING!!!

The survey is now closed. We have a total of 17 persons (18 if I include myself) that indicated their interest in the formation of the AI/ML group.
Thanks for everyone who showed interest. If anyone else is interested in the group, please feel free to jump right into the new thread and introduce yourself. I'll put a link to the new thread here when it's created.

I'll create a thread here on nairaland for the group and will cc (mention) interested person on that thread. We can then discuss on that thread how to proceed and what platform (e.g. nairaland, whatsapp, google group, slack etc) will be preferable.

Recall some important points as stated earlier:
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. We will focus on Modern AI and its application rather than classical AI systems.


Thank you all.

4kings:

That's interesting...
I sure can help, but let me wait for da boss to log in.

Do you understand the concept behind KNN first of all?

I dont fully understand it as a whole, but I know that it has to do with plotting points on a graph. you also plot an unknown point (k) and find its nearest neighbours(usually odd : 3,5 for small data sets). The maximum neighbours matching k is the solution.

E.g. If there are 4 bags in a basket. 3 are branded gucci, versace, and D n G. The fourth is left unclassified. Our problem is to place a brand on the fourth bag.

We first look at the features of each bag - shape, size, leather, color, quality, firmness, tailoring seam. We use this data to plot a graph on an x, y coordinate system (We over simplify the problem set). Then, we get the features of the fourth bag and plot it on the same graph. We assume a number K and draw a line from 4th plot to k. The nearest neighbours along this line tells us which brand bag four belong to. Ideally, the number of bags will be more than four for this to work. I take corrections, please educate!

EbonyMira:


I dont fully understand it as a whole, but I know that it has to do with plotting points on a graph. you also plot an unknown point (k) and find its nearest neighbours(usually odd : 3,5 for small data sets). The maximum neighbours matching k is the solution.

E.g. If there are 4 bags in a basket. 3 are branded gucci, versace, and D n G. The fourth is left unclassified. Our problem is to place a brand on the fourth bag.

We first look at the features of each bag - shape, size, leather, color, quality, firmness, tailoring seam. We use this data to plot a graph on an x, y coordinate system (We over simplify the problem set). Then, we get the features of the fourth bag and plot it on the same graph. We assume a number K and draw a line from 4th plot to k. The nearest neighbours along this line tells us which brand bag four belong to. Ideally, the number of bags will be more than four for this to work. I take corrections, please educate!

Ya, you very close.
Take a look at the quick image i made below.


Assume, the blue, green and yellow are features for Gucci, Versace and D&G respectively.
And the grey point is the test data we need to classify.

Then as you can see from the second image above, the nearest neighbor in the dataset is green(Versace). And K=1 because we are assuming the first nearest neighbor is our classifier.

But when K=3, the data would be classified as yellow(D&G) since it has the highest nearest neighbor.

Finding Nearest Neighbor:
To find closest neighbor or closest distance we can use different distance metrics.
There are many distance metrics such as Cosine Similarity, Euclidean Distance, Mahattan Distance, Minkowski distance and so on.
Finding the right distance metrics to use is an issue on its own so i won't touch that now.

This is the mathematical formula for cosine similarity:


The python implementation is:

import numpy as np
def cosine_similarity(x, y):
	dot_product = np.dot(x, y)
	norm_x = np.linalg.norm(x)
	norm_y = np.linalg.norm(y)
	return dot_product / (norm_x * norm_y)

I'm not sure what programming language you use, but if you use Matlab or Octave then the above should also make sense to you.

Normalization is same as np.sqrt(np.sum(x**2)) or math.sqrt(sum(x**2))
x and y are your feature vectors.

For Euclidean Distance:
This is the formula:


Very simple and is the most used(at least from what I've observed so far).
The python implementation should be:

import numpy as np
def euclidean_distance(x,y):   
    return np.sqrt(np.sum((x-y)**2))

So the steps to write a K-Nearest Neighbor code from scratch are:
1) Get you data-sets X and y. (training features and test data)
2) Choose K. (Number of nearest neighbor to identify)
3) Using the test data write a function that calculates closest distance based on any distance metrics you choose and choose the group it belongs.
----That is get first distance value and also get whether it belongs to yellow, blue or green and continue.
----This depends on how your datasets is arranged, either in a dictionary or sublists or whatever.

4) Sort your result from lowest to highest distance.
5) then slice your result based on K.
----For instance is K =3, and your result is a list variabe. Code will be --> result[:K]

6) Then get the highest number of groups in the new variable.

7) And voila there you go.

I think with these you can implement the code yourself.

---

Parameter Optimization:
- The best value for K is usually determined by cross-validation. That is testing each value for K to get the optimal result.(this is computationally expensive but helpful though)
--- Generally the most used method is k = n^(1/2)

Distance metrics: well this is where statistical skill really come to play.

I didn't want to just write the complete code, but if you don't get the explanation i made while implementing your code you can quote me.

This lecture from MIT should also give you a better explanation.

https://www.youtube.com/watch?v=eg8DJYwdMyg&t=404s

4kings:

Ya, you very close.
Take a look at the quick image i made below.


Assume, the blue, green and yellow are features for Gucci, Versace and D&G respectively.
And the grey point is the test data we need to classify.

Then as you can see from the second image above, the nearest neighbor in the dataset is green(Versace). And K=1 because we are assuming the first nearest neighbor is our classifier.

But when K=3, the data would be classified as yellow(D&G) since it has the highest nearest neighbor.

Finding Nearest Neighbor:
To find closest neighbor or closest distance we can use different distance metrics.
There are many distance metrics such as Cosine Similarity, Euclidean Distance, Mahattan Distance, Minkowski distance and so on.
Finding the right distance metrics to use is an issue on its own so i won't touch that now.

This is the mathematical formula for cosine similarity:


The python implementation is:

import numpy as np
def cosine_similarity(x, y):
	dot_product = np.dot(x, y)
	norm_x = np.linalg.norm(x)
	norm_y = np.linalg.norm(y)
	return dot_product / (norm_x * norm_y)

I'm not sure what programming language you use, but if you use Matlab or Octave then the above should also make sense to you.

Normalization is same as np.sqrt(np.sum(x**2)) or math.sqrt(sum(x**2))
x and y are your feature vectors.

For Euclidean Distance:
This is the formula:


Very simple and is the most used(at least from what I've observed so far).
The python implementation should be:

import numpy as np
def euclidean_distance(x,y):   
    return np.sqrt(np.sum((x-y)**2))

So the steps to write a K-Nearest Neighbor code from scratch are:
1) Get you data-sets X and y. (training features and test data)
2) Choose K. (Number of nearest neighbor to identify)
3) Using the test data write a function that calculates closest distance based on any distance metrics you choose and choose the group it belongs.
----That is get first distance value and also get whether it belongs to yellow, blue or green and continue.
----This depends on how your datasets is arranged, either in a dictionary or sublists or whatever.

4) Sort your result from lowest to highest distance.
5) then slice your result based on K.
----For instance is K =3, and your result is a list variabe. Code will be --> result[:K]

6) Then get the highest number of groups in the new variable.

7) And voila there you go.

I think with these you can implement the code yourself.

---

Parameter Optimization:
- The best value for K is usually determined by cross-validation. That is testing each value for K to get the optimal result.(this is computationally expensive but helpful though)
--- Generally the most used method is k = n^(1/2)

Distance metrics: well this is where statistical skill really come to play.

I didn't want to just write the complete code, but if you don't get the explanation i made while implementing your code you can quote me.

This lecture from MIT should also give you a better explanation.

https://www.youtube.com/watch?v=eg8DJYwdMyg&t=404s

Thanks for the explanation. I am familiar with python, but I currently work with Javascript - nodejs. I hope we would be able to work together.

EbonyMira:


Thanks for the explanation. I am familiar with python, but I currently work with Javascript - nodejs. I hope we would be able to work together.

Okey-dokey.
We all would work together, that's the vision of SoftEng.

Hey guys, I'm interested in machine learning but I have zero knowledge apart from its definition.

@SoftEng. Now what next?

Unmarshalled:
Hey guys, I'm interested in machine learning but I have zero knowledge apart from its definition.

Let's wait for SoftEng.

You can start with some courses from MIT OpenCourseWare.

4kings:

Let's wait for SoftEng.

You can start with some courses from MIT OpenCourseWare.

OK tanx

@4kings, @Unmarshalled and @everyone. I apologise for the delays.

Since we have more than 10 persons interested in the group, I'll go ahead an create a thread now on Nairaland as earlier stated.

@4kings thanks for your inputs so far.

I am also interested.

SoftEng:
@4kings, @Unmarshalled and @everyone. I apologise for the delays.

Since we have more than 10 persons interested in the group, I'll go ahead an create a thread now on Nairaland as earlier stated.

@4kings thanks for your inputs so far.

Atlast.

Waiting for d thread.. I'm thrilled

The new thread for the group has now been created.
Visit here https://www.nairaland.com/4237475/artificial-intelligence-machine-learning-group.

This survey thread is now closed.

Thank you.

Interested

Im interested

odizeey:
Im interested

Here's the link to contribute to: https://www.nairaland.com/4237475/artificial-intelligence-machine-learning-group

https://www.youtube.com/watch?v=RzG8LQ4Ntuw

Data Science Nigeria Opens First Nigerian Artificial Intelligence HUB in UNILAG

Data Science Nigeria in furtherance of its drive to boost research and innovation in Nigeria, opened the first ever
Nigerian Artificial Intelligence Hub in the University of Lagos.

AI-infused mobile apps give marketers a complete understanding of user behavior and help refine the entire journey by delivering value at every step of the way.

It sifts through a huge number of data points to isolate actionable insights that’ll pave the way for a more meaningful relationship with the customers.

https:///BTuje4

I'm interested...

interested

am interested

I am very much interested