SoftEng:
Hi all,

Here's another competition that might be of interest to you.

Microsoft Cloud AI Research Challenge.
https://www.microsoft.com/en-us/research/academic-program/microsoft-cloud-ai-research-challenge/

The Challenge
Use Microsoft AI services to develop (breakthrough) applications or embed their AI (cognitive) services into your application.
Prize: 25,000 USD Grand Prize

My take on the challenge
In simple terms, Microsoft wants to encourage more developers to use their AI platform and also discover new use cases of AI from these challenge. I believe it is a fun and interesting challenge. For anyone who will be interested, you can think of wide variety of application areas, but if you can narrow it down to application areas that solves some of our local problems that would be a good thing.

Thanks.

4kings:
Darivie04 welcome!!!
It's very cool that you are in JSS3 and are tackling machine learning. That's dope.

Anyways, a while back i spoke with raymod170 offline about taking fundamentals of machine learning.
That is explaining the maths behind most algorithms in simple details instead of using libraries without understanding what's going on under the hood as that impedes the capability to understand and make research in this field.

I've been busy, but would put it in my schedule to explain the details behind most machine learning algorithms in simple details including reinforcement learning, occasionally on this thread.

osarenomaspecial:

taking fundamentals of machine learning will be great at this point is what we need now in the thread.

4kings:
Great!!!

4kings:

"Unifying theory of data mining"?
I don't see how that's possible for now.
At the end of the day it will still be selection of best-fit algorithm on some set of categories which might even have a high bias variance based on the dataset.

Besides you are talking of unified theory and also talking on comparing two algorithms, i'm lost.

You can make your research domain based instead of data mining in general or maybe derive hierarchical clusters as "unifying theory" for cluster selection in any dataset(which still isn't as efficient when comparing with Inpedndent component analysis(ICA) on specific problem set).
That's as close as i can think.

Anyways do tell about the outcome of your research or any progress you make.

MYSELF2018:
Thanks a lot my brother

Well here i start :

#This is a python library for plotting of graphs.
from matplotlib import pyplot as plt

#Make variables for our dataset with their values ordered appropriately
time =   [10,  20, 30, 40, 50, 60]
wages = [75000, 150000, 225000, 300000, 375000, 450000]

#Code for plotting the graph of the dataset.
plt.plot(time, wages) #Use plt.scatter to see the scatter plot instead.
plt.yticks(wages)
plt.xlabel('Time(in Hours)')
plt.ylabel('Wages(Naira)')
plt.show()


#define function for the linear equation (y = mx + b)
#The linear equation function should take additional argument x to be used for prediction.
#data1 and data2 represents time and wages.
def linear_equation(data1, data2, x):
    #remember how to derive m.
    #here we used the last two points on the x and y axes.
    # (y2 - y1)/(x2 - x1)
    m = (data2[-1] - data2[-2]) / (data1[-1] - data1[-2])
    
    #Remember how to get b, by substituting the corresponding values for a point(x,y).
    #Use the last values for x and y also for simplicity sake.
    #b = y - mx by substitution.
    b = data2[-1] - (m * data1[-1])
    
    #therefore to predict y given the argument x:
    y = (m * x) + b
    return y

#Like in the explanation:
#To predict wage(y) when time(x) is 70:
#Call the function with the argument for x also putting the two intial arguments for the dataset.
predict = 70
result = linear_equation(time, wages, predict)
print("The wage prediction when time is {} is:".format(70), result)

Samcent:
This is great! Following the lesson....

Darivie04:
@4kings what your doing is so informative I really appreciate it!!

omohayek:

You do? Alright then, what is Q-learning? What is partial observability? Can you tell me what the Bellman equation is, and how it works? Please provide the answers to these questions in your own words, instead of regurgitating whatever you can find on Google.

If you have the book, I doubt you've even begun to read it, as otherwise you wouldn't be on here asking such rudimentary questions.

agwaisrael:


Typical Nigerian, scratch that. Typical Hater!

It doesn't promote progress trying to find flaws, but instead try to build.

What point are you trying to prove? If you are so much of an expert, why haven't you created your own algorithm or even a common ml library.

Try not to look for flaws in people and if you do find, don't be a hater. Find strengths

omohayek:

God, another mouth-breather! Didn't you read the part where I already said I get paid to build full time? And I earn far more doing so than a sub-literate bozo like you ever will, that much is certain.


Except, to repeat myself, I have in fact done such things, not that you'd know the significance of any of them if I bothered to mention them to you. You wouldn't know the Bellman equation from Bell's inequality if your right to keep living depended on the answer, yet you dare to presume to know what I have or haven't done? By God you're dumb! Here, tell us all the right answer to this simple question and I'll deign to spend some time telling you something about what I've achieved and been well-rewarded for: how many neurons in a feed-forward architecture would it take to master pendulum-balancing in the OpenAI Gym environment, and what sort of non-linear activations would you use to accomplish it? Bonus points for sharing with all of Nairaland what optimization technique you'd use, and what hyperparameters you'd employ. Somehow, I suspect we'll see Halley's Comet return before you can answer even such elementary questions.


Save your utterly stupid, content-free pontifications for someone of your own dishwater-level of intellectual depth. It's not compulsory to comment when you have nothing of value to add to a discussion. I know it's hard for half-wits like you to understand, but not everyone in the world is on your low level of bare-survival doing whatever inconsequential menial tasks you do to survive (assuming you even have a job of any kind).

omohayek:

Save your utterly stupid, content-free pontifications for someone of your own dishwater-level of intellectual depth. It's not compulsory to comment when you have nothing of value to add to a discussion. I know it's hard for half-wits like you to understand, but not everyone in the world is on your low level of bare-survival doing whatever inconsequential menial tasks you do to survive (assuming you even have a job of any kind).

agwaisrael:


Typical Nigerian, scratch that. Typical Hater!

It doesn't promote progress trying to find flaws, but instead try to build.

What point are you trying to prove? If you are so much of an expert, why haven't you created your own algorithm or even a common ml library.

Try not to look for flaws in people and if you do find, don't be a hater. Find strengths

Damoxy:
Good Day everyone, just joined this Group via someone here. I really want to learn Data Science. I dont know how it connects to AI or ML. If it does. How then can I start learning it.

The softwares and books and books that I will need to kick start this learning process.

Darivie04:


https://www.google.com.ng/url?q=http://bigdata-madesimple.com/8-best-python-data-science-books/&sa=U&ved=2ahUKEwid8Oa9lqraAhUDKFAKHUFrCsIQFjAAegQICRAB&usg=AOvVaw0kTU8N0d341QHj7JPB0GJw
I personally uses the book called Data Science From Scratch
And I guess Data Science is connected to ML and AI cause you have to prepare the Data you will use for training in ML

zamie:


Mr Man, YOU ARE FOOLISH. Fools talk like this. Someone is trying to give you an advice and you are here spitting English. Because, it's u that started machine learning abi. Pride Comes before a Fall.

Just get off this Page, We don't need people like you who know nothing and yet make noise. Prove me wrong and tell me one achievement you have made. Instead, you will come up with more English. Incase u don't know, Machine learning is for smart people unlike people like u with an empty skull but a big mouth.

Let me give you a simple Question to even know ur level of intelligence sef.

What is 1 + 1?
(pls, no calculators allowed)

OLODO, ANSWER THIS QUESTION O, DON'T COME AND BE SPEAKING ENGLISH

4kings:
Hey guys, been busy would continue the tutorial maybe this night when i get home.

By the way my images are not showing again, abi is it my laptop.

Can anyone recommend a good image sharing site, i'm using imgur and apparently it's no good.

Darivie04:

So whats today's topic?

4kings:

Today!!!
We would be looking at the inefficiency of the linear equation on a non evenly distributed dataset, therefore introducing the concept of optimisation or error functions and a look at multiple linear regression, then we'd write code for predicting stock market data.
Insha Allah...

 
from matplotlib import pyplot as plt
prices = [100, 100, 200, 200, 300, 300, 400, 400, 500, 500, 600, 600, 700, 700, 800]
hours = [1,2,3,4,5,6,7,8,9,10,11,12, 13, 14, 15]
plt.xticks(hours) #To indicate the numbers on x axis.
plt.scatter(hours, prices)  

#Code:
def linear_equation(data1, data2, x):
    #Average of y divided by Average of X
    m = (sum(data2)/len(data2)) / (sum(data1)/len(data1))
    
    #Remember how to get b, by substituting the corresponding values for a point(x,y).
    #Use the last values for x and y also for simplicity sake.
    #b = y - mx by substitution.
    b = data2[-1] - (m * data1[-1])
    
    #therefore to predict y given the argument x:
    y = (m * x) + b
    return y

#Code:
plt.xticks(hours)
plt.scatter(hours, prices)

line_y = [linear_equation(hours, prices, i) for i in hours] #Draw line by connecting all values for y. 
plt.plot(hours, line_y)
plt.show()

#Code:
predictions = [16, 20]
result = [linear_regression(hours, prices, i) for i in predictions]
print(result)

#Code:
def linear_regression(x,y, predict):
    #Calcluating m
    x_mean = sum(x)/len(x)
    y_mean = sum(y)/len(y)
    
    
    numerator = 0
    for i in range(0, len(x)):
        numerator += (x[i] - x_mean) * (y[i] - y_mean)
        
    denominator = 0
    for i in range(0, len(x)):
        denominator += (x[i] - x_mean) ** 2
        
    m = numerator/denominator
    
    #Now let's calculate b
    b = y_mean - (m * x_mean)
    
    #So to predict y:
    #y = mx + b
    y = (m * predict) + b
    return y

#Code:
import numpy
y = np.array([1,2,3,4,5])
X = np.array([[2,4,6,8,10], [3,6,9,12,15], [4,8,12,16,20]])

X = X.T #Get X transpose.
coef = np.linalg.lstsq(X,y)[0] #This function follows the linear matrix equation.

features = [12, 18, 24]
prediction = np.dot(features, coef)
print(prediction)
>>> 6

#Code:
#Remember to convert the variales to numpy arrays instead of python lists.
def LinearRegression(X,y,predict):
    #x and y would be in numpy arrays.
        
    X = X.T
    coef = np.linalg.lstsq(X,y)[0]
    
    result = np.dot(predict, coef)
    return result