Table of Contents
This is to classify any types of fruit into good or bad by first learning what a good fruit is and what a bad one is based on 3 features that the user trains with, Taste, edibility of Seeds and edibility of the Skin. This maybe useless since all popular languages will come with libraries that will create thousands of perceptrons in a single line but this is useful for a full understand of how one works.
This documentation will describe how the Python Version of the code is being run but you can apply the same logic to the R and C version as well.
This section should list any major frameworks/libraries used to bootstrap your project. Leave any add-ons/plugins for the acknowledgements section. Here are a few examples.
You will need to have the proper IDE / environments to run languages, no libraries were used so this code will be compatible with most versions of the language.
I have used the below IDEs for the languages
- R : R Studio
- C : Command Line
- Python : Command Line
Running the script is very straight forward, just run the corresponding script, Example : python3 fruit.c or ./fruit.sh after compiling with C
- We first initialize a single neutron with three inputs, a single output. All the weights are initially set to zero. The algorithm used as the activation function is
if sum(weight1, weight2, weight3) > 0.4:
Fruit
else:
Not a Fruit
You can visualize the above neutron as below
We train the neutron with a few known classes. First we train with a fruit having good taste, edible seeds and edible skin. So the input would be (1,1,0) as below
We feed this input the code
while j<=3:
print("The weights are "+str(w[1])+" "+str(w[2])+" "+str(w[3]))
x[i][j]=int(input("Enter the qualities of the fruit\n"))
print("The entered number is "+str(x[i][j]))
y = y + (w[j] * x[i][j])
if y>0.4:
y2=1
else:
y2=0
j=j+1
And we will get the output as 0 or it's not a fruit. We know that it is a fruit and whenever the predcited and actual values are different we run the correction part
if(y2!=yl):
j=1
while j<=3:
print("\nThe weight is :"+str(w[j]))
dw[j] = (yl - y2) * x[i][j] * l
print("l="+str(l)+"\n")
w[j] = w[j] + dw[j]
print("Now the weight is :"+str(w[j]))
j=j+1
else:
print("The output of the computer and user matched we go on to the next fruit\n")
We now use the Perceptron Learning Rule to correct our mistake
∆w = learning rate X (teacher provided class – predicted class) X node input
Here we use a
1. learning rate = 0.25
2. teacher provided class – predicted class of 1
3. node input = 1 ( for the first node, 1 for the second and 0 for the third )
Plugging this in we get a value for ∆w = 0.25 X 1 X 1 = 0.25
∆w tells us how much we should change the respective nodes of the perceptron. In our case we will change the first node weight to 0.25 using the above result. Similarly, all the weights are changed for the respective input. And finally it would look something like this
We do the above as many times as want, the general rule is the more we train the better out model will get.
After training for 10 train fruits the weights are 0.5, 0.25 and 0.25
Now comes the testing phase where we once again show the perceptron 3 features of a fruit
x1=input()
x2=input()
x3=input()
y = w[1] * x1 + w[2] * x2 + w[3] * x3
if y>0.4:
print("The answer is one hence the fruit is good\n")
else:
print("The answer is zero hence the fruit is bad\n")
We now test the perceptron with features of a fruit it has never seen i.e., never has been trained to indentify
Now with all the new weights the perceptron is able to accuratly detect the fruit, even though it has never seen this.
Distributed under the MIT License. See LICENSE.txt for more information.
Your Name - Anandha Krishnan H - [email protected]
Project Link: https://github.com/anandhakrishnanh/Fruit-Classifier-Perceptron