Scribd
Upload
182 views35 pages

Build Neural Networks with Python AI

The document explains how to build a basic neural network from scratch in Python by implementing the necessary components like neurons, weights, and forward propagation. It also covers computing predictions from the network and calculating error using mean squared error to evaluate accuracy.

Uploaded by

makanakili
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
182 views35 pages

Build Neural Networks with Python AI

The document explains how to build a basic neural network from scratch in Python by implementing the necessary components like neurons, weights, and forward propagation. It also covers computing predictions from the network and calculating error using mean squared error to evaluate accuracy.

Uploaded by

makanakili
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd

Python AI: How to Build a Neural Network & Make Predictions – Rea... [Link]

com/python-ai-neural-network/

1 of 35 3/19/2021, 1:08 PM
Python AI: How to Build a Neural Network & Make Predictions – Rea... [Link]

2 of 35 3/19/2021, 1:08 PM
Python AI: How to Build a Neural Network & Make Predictions – Rea... [Link]

3 of 35 3/19/2021, 1:08 PM
Python AI: How to Build a Neural Network & Make Predictions – Rea... [Link]

4 of 35 3/19/2021, 1:08 PM
Python AI: How to Build a Neural Network & Make Predictions – Rea... [Link]

5 of 35 3/19/2021, 1:08 PM
Python AI: How to Build a Neural Network & Make Predictions – Rea... [Link]

6 of 35 3/19/2021, 1:08 PM
Python AI: How to Build a Neural Network & Make Predictions – Rea... [Link]

7 of 35 3/19/2021, 1:08 PM
Python AI: How to Build a Neural Network & Make Predictions – Rea... [Link]

8 of 35 3/19/2021, 1:08 PM
Python AI: How to Build a Neural Network & Make Predictions – Rea... [Link]

price = (weights_area * area) + (weights_age * age) + bias

weights_area weights_age

9 of 35 3/19/2021, 1:08 PM
Python AI: How to Build a Neural Network & Make Predictions – Rea... [Link]

weights_2

input_vector weights_1

input_vector weights_1

input_vector weights_2

10 of 35 3/19/2021, 1:08 PM
Python AI: How to Build a Neural Network & Make Predictions – Rea... [Link]

venv

$ python ‐m venv ~/.my‐env
$ source ~/.my‐env/bin/activate

pip

(my‐env) $ python ‐m pip install ipython numpy matplotlib
(my‐env) $ ipython

input_ve

In [1]: input_vector = [1.72, 1.23]


In [2]: weights_1 = [1.26, 0]
In [3]: weights_2 = [2.17, 0.32]

In [4]: # Computing the dot product of input_vector and weights_1
In [5]: first_indexes_mult = input_vector[0] * weights_1[0]
In [6]: second_indexes_mult = input_vector[1] * weights_1[1]
In [7]: dot_product_1 = first_indexes_mult + second_indexes_mult

In [8]: print(f"The dot product is: {dot_product_1}")
Out[8]: The dot product is: 2.1672

2.1672
dot_product_1 [Link]()

In [9]: import numpy as np

In [10]: dot_product_1 = [Link](input_vector, weights_1)

In [11]: print(f"The dot product is: {dot_product_1}")
Out[11]: The dot product is: 2.1672

11 of 35 3/19/2021, 1:08 PM
Python AI: How to Build a Neural Network & Make Predictions – Rea... [Link]

[Link]()
input_vector weights_2

In [10]: dot_product_2 = [Link](input_vector, weights_2)

In [11]: print(f"The dot product is: {dot_product_2}")
Out[11]: The dot product is: 4.1259

4.1259
0
0

0
input_vector weights_2 4.1259 4.1259 2.1672 input_

12 of 35 3/19/2021, 1:08 PM
Python AI: How to Build a Neural Network & Make Predictions – Rea... [Link]

0 1

[Link](x)

0 1

0 1
1 0.5 0

13 of 35 3/19/2021, 1:08 PM
Python AI: How to Build a Neural Network & Make Predictions – Rea... [Link]

14 of 35 3/19/2021, 1:08 PM
Python AI: How to Build a Neural Network & Make Predictions – Rea... [Link]

In [12]: # Wrapping the vectors in NumPy arrays
In [13]: input_vector = [Link]([1.66, 1.56])
In [14]: weights_1 = [Link]([1.45, ‐0.66])
In [15]: bias = [Link]([0.0])

In [16]: def sigmoid(x):
   ...:  return 1 / (1 + [Link](‐x))

In [17]: def make_prediction(input_vector, weights, bias):


   ...:  layer_1 = [Link](input_vector, weights) + bias
   ...:  layer_2 = sigmoid(layer_1)
   ...:  return layer_2

In [18]: prediction = make_prediction(input_vector, weights_1, bias)

In [19]: print(f"The prediction result is: {prediction}")
Out[19]: The prediction result is: [0.7985731]

0.79 0.5 1
[Link]([2, 1.5]) 0

In [20]: # Changing the value of input_vector
In [21]: input_vector = [Link]([2, 1.5])

In [22]: prediction = make_prediction(input_vector, weights_1, bias)

In [23]: print(f"The prediction result is: {prediction}")
Out[23]: The prediction result is: [0.87101915]

0.5
0.87

15 of 35 3/19/2021, 1:08 PM
Python AI: How to Build a Neural Network & Make Predictions – Rea... [Link]

In [24]: target = 0

In [25]: mse = [Link](prediction ‐ target)

In [26]: print(f"Prediction: {prediction}; Error: {mse}")
Out[26]: Prediction: [0.87101915]; Error: [0.7586743596667225]

0.75

16 of 35 3/19/2021, 1:08 PM
Python AI: How to Build a Neural Network & Make Predictions – Rea... [Link]

layer_1

error = [Link](prediction ‐ target) (predi
x error = [Link](x)

A
B

17 of 35 3/19/2021, 1:08 PM
Python AI: How to Build a Neural Network & Make Predictions – Rea... [Link]

ⁿ ⁽ⁿ⁻ ⁾
x 1

error = [Link](prediction ‐ target)
x 2 * x
x error

weights_1
0.75 2 * x
2

In [27]: derivative = 2 * (prediction ‐ target)

In [28]: print(f"The derivative is {derivative}")
Out[28]: The derivative is: [1.7420383]

1.74
weights_1

In [29]: # Updating the weights
In [30]: weights_1 = weights_1 ‐ derivative

In [31]: prediction = make_prediction(input_vector, weights_1, bias)

In [32]: error = (prediction ‐ target) ** 2

In [33]: print(f"Prediction: {prediction}; Error: {error}")
Out[33]: Prediction: [0.01496248]; Error: [0.00022388]

18 of 35 3/19/2021, 1:08 PM
Python AI: How to Build a Neural Network & Make Predictions – Rea... [Link]

A B

0.1 0.01 0.001

[Link](x) x

weights_1 bias

19 of 35 3/19/2021, 1:08 PM
Python AI: How to Build a Neural Network & Make Predictions – Rea... [Link]

20 of 35 3/19/2021, 1:08 PM
Python AI: How to Build a Neural Network & Make Predictions – Rea... [Link]

derror_dweights

derror_dweights

derror_dweights = (
derror_dprediction * dprediction_dlayer1 * dlayer1_dweights
)

y = f(x) f x
derror_dprediction

derror_dbias
bias

derror_dprediction

21 of 35 3/19/2021, 1:08 PM
Python AI: How to Build a Neural Network & Make Predictions – Rea... [Link]

derror_dprediction
dprediction_dlayer1

1 ‐ sigmoid(x)

22 of 35 3/19/2021, 1:08 PM
Python AI: How to Build a Neural Network & Make Predictions – Rea... [Link]

layer_1
1
derror_dbias

In [36]: def sigmoid_deriv(x):
   ...:  return sigmoid(x) * (1‐sigmoid(x))

In [37]: derror_dprediction = 2 * (prediction ‐ target)


In [38]: layer_1 = [Link](input_vector, weights_1) + bias
In [39]: dprediction_dlayer1 = sigmoid_deriv(layer_1)
In [40]: dlayer1_dbias = 1

In [41]: derror_dbias = (
   ...:  derror_dprediction * dprediction_dlayer1 * dlayer1_dbias
   ...: )

NeuralNetwork learning_rate
_compute_derivatives() _update_parameters()
NeuralNetwork

23 of 35 3/19/2021, 1:08 PM
Python AI: How to Build a Neural Network & Make Predictions – Rea... [Link]

class NeuralNetwork:
def __init__(self, learning_rate):
[Link] = [Link]([[Link](), [Link]()])
[Link] = [Link]()
self.learning_rate = learning_rate

def _sigmoid(self, x):


return 1 / (1 + [Link](‐x))

def _sigmoid_deriv(self, x):


return self._sigmoid(x) * (1 ‐ self._sigmoid(x))

def predict(self, input_vector):


layer_1 = [Link](input_vector, [Link]) + [Link]
layer_2 = self._sigmoid(layer_1)
prediction = layer_2
return prediction

def _compute_gradients(self, input_vector, target):


layer_1 = [Link](input_vector, [Link]) + [Link]
layer_2 = self._sigmoid(layer_1)
prediction = layer_2

derror_dprediction = 2 * (prediction ‐ target)


dprediction_dlayer1 = self._sigmoid_deriv(layer_1)
dlayer1_dbias = 1
dlayer1_dweights = (0 * [Link]) + (1 * input_vector)

derror_dbias = (
derror_dprediction * dprediction_dlayer1 * dlayer1_dbias
)
derror_dweights = (
derror_dprediction * dprediction_dlayer1 * dlayer1_dweights
)

return derror_dbias, derror_dweights

def _update_parameters(self, derror_dbias, derror_dweights):


[Link] = [Link] ‐ (derror_dbias * self.learning_rate)
[Link] = [Link] ‐ (
derror_dweights * self.learning_rate
)

24 of 35 3/19/2021, 1:08 PM
Python AI: How to Build a Neural Network & Make Predictions – Rea... [Link]

NeuralNetwork()

In [42]: learning_rate = 0.1

In [43]: neural_network = NeuralNetwork(learning_rate)

In [44]: neural_network.predict(input_vector)
Out[44]: array([0.79412963])

train() NeuralNetwork

train()

25 of 35 3/19/2021, 1:08 PM
Python AI: How to Build a Neural Network & Make Predictions – Rea... [Link]

class NeuralNetwork:
# ...

def train(self, input_vectors, targets, iterations):


cumulative_errors = []
for current_iteration in range(iterations):
# Pick a data instance at random
random_data_index = [Link](len(input_vectors))

input_vector = input_vectors[random_data_index]
target = targets[random_data_index]

# Compute the gradients and update the weights
derror_dbias, derror_dweights = self._compute_gradients(
input_vector, target
)

self._update_parameters(derror_dbias, derror_dweights)

# Measure the cumulative error for all the instances
if current_iteration % 100 == 0:
cumulative_error = 0
# Loop through all the instances to measure the error
for data_instance_index in range(len(input_vectors)):
data_point = input_vectors[data_instance_index]
target = targets[data_instance_index]

prediction = [Link](data_point)
error = [Link](prediction ‐ target)

cumulative_error = cumulative_error + error


cumulative_errors.append(cumulative_error)

return cumulative_errors

_compute_gradients()

update parameters()

26 of 35 3/19/2021, 1:08 PM
Python AI: How to Build a Neural Network & Make Predictions – Rea... [Link]

100

prediction

error

cumulative_error

NeuralNetwork

input_vect

27 of 35 3/19/2021, 1:08 PM
Python AI: How to Build a Neural Network & Make Predictions – Rea... [Link]

In [45]: # Paste the NeuralNetwork class code here
   ...: # (and don't forget to add the train method to the class)

In [46]: import [Link] as plt

In [47]: input_vectors = [Link](
   ...:  [
   ...:  [3, 1.5],
   ...:  [2, 1],
   ...:  [4, 1.5],
   ...:  [3, 4],
   ...:  [3.5, 0.5],
   ...:  [2, 0.5],
   ...:  [5.5, 1],
   ...:  [1, 1],
   ...:  ]
   ...: )

In [48]: targets = [Link]([0, 1, 0, 1, 0, 1, 1, 0])

In [49]: learning_rate = 0.1

In [50]: neural_network = NeuralNetwork(learning_rate)

In [51]: training_error = neural_network.train(input_vectors, targets, 10000)

In [52]: [Link](training_error)
In [53]: [Link]("Iterations")
In [54]: [Link]("Error for all training instances")
In [54]: [Link]("cumulative_error.png")

NeuralNetwork train() input_vectors


10000

28 of 35 3/19/2021, 1:08 PM
Python AI: How to Build a Neural Network & Make Predictions – Rea... [Link]

29 of 35 3/19/2021, 1:08 PM
Python AI: How to Build a Neural Network & Make Predictions – Rea... [Link]

30 of 35 3/19/2021, 1:08 PM
Python AI: How to Build a Neural Network & Make Predictions – Rea... [Link]

31 of 35 3/19/2021, 1:08 PM
Python AI: How to Build a Neural Network & Make Predictions – Rea... [Link]

32 of 35 3/19/2021, 1:08 PM
Python AI: How to Build a Neural Network & Make Predictions – Rea... [Link]

33 of 35 3/19/2021, 1:08 PM
Python AI: How to Build a Neural Network & Make Predictions – Rea... [Link]

34 of 35 3/19/2021, 1:08 PM
Python AI: How to Build a Neural Network & Make Predictions – Rea... [Link]

35 of 35 3/19/2021, 1:08 PM

You might also like