Simplest Neural Network (with Keras)

In this tutorial, we will learn how to write our own neural network.
We will write the simplest possible neural network. It will have 1 layer with 1 neuron.

Look at the following patterns:
x => -1, 0, 1, 2, 3, 4
y => -3, -1, 1, 3, 5, 7

Can you guess the value of y for a given value of x ?

Did you guess it ?

What do you think the value of y will be if x = 10 ?

Do you think our Neural Network can guess it ?

Let’s see
For complete code: Click here

Import the libraries

import tensorflow as tf
import numpy as np

Tensorflow allows us to build neural networks in our Python program.
Numpy allows us to create matrices very easily in our Python program.

Structure of the program


Let us discuss the basic structure of writing a deep learning program.

Step 1: Collect the training data

Collect as much training data as possible. These examples of correct X and Y, will help our neural network to learn the pattern.

In this case, we have our training data of 6 examples of values of X and Y.

x = np.array([-1.0, 0.0, 1.0, 2.0, 3.0, 4.0])
y = np.array([-3.0, -1.0, 1.0, 3.0, 5.0, 7.0])

Step 2: Build the neural network

Build the neural network model: We will decide on how many layers and how many neurons in each layer.

model = tf.keras.Sequential()
model.add(tf.keras.layers.Dense(units=1, input_shape=[1]))

Step 3: Compile the model

We will decide the optimizer algorithm and also the loss function to use for our neural network model.

model.compile(optimizer='sgd', loss='mean_squared_error')

Step 4: Feed the training data

The neural network tries to find a function that fits the given data. We will decide for how many epochs (iterations) our neural network will the learning process.

1, y, epochs=500)

Step 5: Test the model

We show the neural network, examples of data outside the training data. The neural network must guess the output for new data as accurately as possibleMake some predictions

values = np.array([10.0])

predictions = model.predict(values)