Tutorial 1:
Building a Perceptron
Objectives:
• Building Perceptron from Scratch
• How Weights and Bias get Updated
• Activation Function
• Data Loading
• Training & Testing
Lets build a Perceptron?
Step 1 — Perceptron Class
Follow these simple steps:
• Create a file called “Perceptron.py”
• Define a class called Perceptron inside
To build a perceptron, we need 3 attributes:
• η (eta): the learning rate is usually a small value between 0.0 and 1.0 which
defines how quickly the model learns.
• n_iter: the number of iterations.
• w_: the weights. A weight defines the importance of the corresponding
input value
Step 2 — Weighted sum
np.dot(X, w_) is simply:
Step 3 — Step function
The step function predict () returns:
• 1 — if the weighted sum is greater than 0.0
• -1 — Otherwise.
Step 4— Create the training method
An explanation will follow.
We initialize the weights to 0, then start our training.
For each input:
1. we calculate ŷ, the predicted output.
2. then we update the weights from W1 to Wn (remember Wo is the bias).
We repeat this process n_iter times (n_iter is the number of iterations).
Weights update in the perceptron
How are the weights updated?
Every weight Wi is updated by Δ(Wi).
Wi = Wi + Δ(Wi)
Where:
Δ(Wi) = η (y — ŷ) * Xi
Don’t panic, it’s easier than it seems:
• Wi is i-th weight.
• Δ(Wi) is the update. It determines if the weight increases or decreases.
• η is the learning rate.
• y is the actual value.
• ŷ is the predicted value.
• Xi is the i-th input.
Example
Get your pen, we are going to do some math.
Random data to calculate weights update
Δ(Wi) = η (y — ŷ) * Xi
When y is equal to ŷ, there is no update:
• Δ(W1) = 0.01 * ( 1 − 1 ) ∗ 0.5 = 0
• Δ(W2) = 0.01 * ( (−1) − (−1) ) ∗ 1 = 0
When y is different than ŷ, it means that there’s an error and the weights must be
updated by Δ(Wi):
• Δ(W3) = 0.01 * ( 1 − (−1) ) ∗ 4 = 0.08
• Δ(W4) = 0.01 * ( (−1) − 1) ∗ 2 = −0.04
Whew!
We made it through to the end, let’s test our Perceptron.
Step 5 : Loading Data:
The data is loaded from https://archive.ics.uci.edu/ml/machine-learning-
databases/iris/iris.data it represents the Iris dataset. It is a famous dataset used
for classification tasks. It contains 150 samples of iris flowers, each described by
four features (sepal length, sepal width, petal length, and petal width) and a
target label that indicates the species of the flower (either Iris-setosa, Iris-
versicolor, or Iris-virginica).
Output
Separating the data (X) from the labels (y)
Splitting the data into train and test set
Output: ( The output might Appear different on your screens )
Step 6: Training the perceptron:
Step 7 : Making predictions on test data
Step 8: Measuring Performances
Tasks:
1. Take manual input for the four features (sepal length, sepal width,
petal length, and petal width).
▪ Use the model to predict whether it's Iris-setosa or not based
on the manually provided values.
▪ Display the prediction result.
2. Replace the Step Function with Sigmoid.
3. Replace the labels of 1,-1 with 1 & 0.
