python-tensorflowHow do I create a simple neural network using Python and TensorFlow?

Creating a simple neural network using Python and TensorFlow is relatively straightforward. To begin, you'll need to import the necessary packages:

import tensorflow as tf
import numpy as np

Next, you'll need to define the model's architecture. This can be done by creating layers of neurons, each of which will have an associated weight and bias value. The following example creates a single layer with three neurons:

model = tf.keras.Sequential([
    tf.keras.layers.Dense(3, activation='relu', input_shape=(1,))
])

Once the model architecture is defined, you'll need to compile it with an optimizer and a loss function. The following example uses the Adam optimizer and a mean squared error loss function:

model.compile(optimizer='adam',
              loss='mean_squared_error',
              metrics=['accuracy'])

Finally, you'll need to train the model on a set of data. The following example uses a NumPy array of data to train the model:

X = np.array([[1], [2], [3], [4]])
y = np.array([[2], [4], [6], [8]])
model.fit(X, y, epochs=1000)

The output of the above code should look something like this:

Epoch 1/1000
1/1 [==============================] - 0s 1ms/step - loss: 0.0013 - accuracy: 0.0000e+00
Epoch 1000/1000
1/1 [==============================] - 0s 2ms/step - loss: 2.9585e-05 - accuracy: 0.0000e+00

Once the model is trained, you can use it to make predictions on new data.

Code explanation

• import tensorflow as tf - imports the TensorFlow library
• import numpy as np - imports the NumPy library
• tf.keras.Sequential([]) - creates a model architecture with a single layer of neurons
• tf.keras.layers.Dense(3, activation='relu', input_shape=(1,)) - creates a layer of three neurons with a rectified linear activation function and an input shape of 1
• model.compile(optimizer='adam', loss='mean_squared_error', metrics=['accuracy']) - compiles the model with an Adam optimizer and a mean squared error loss function
• X = np.array([[1], [2], [3], [4]]) - creates a NumPy array with input data
• y = np.array([[2], [4], [6], [8]]) - creates a NumPy array with output data
• model.fit(X, y, epochs=1000) - trains the model on the input and output data for 1000 epochs

Helpful links

• TensorFlow Documentation
• Creating a Simple Neural Network in TensorFlow

Edit this code on GitHub