This page presents a neural network curve fitting example. This example shows and details how to create nonlinear regression with TensorFlow.
The following has been performed with the following version:
Try the example online on Google Colaboratory.
The goal of this example is to approximate a nonlinear function given by the following equation:
$$ y = 0.1.x.\cos(x) $$
The blue dots are the training set, the red line is the output of the network:
Each line is explained in the next section. Source code and example can be run online on Google Colaboratory
First, we import the libraries:
import numpy as np
import matplotlib.pyplot as plt
from tensorflow import keras
from google.colab import files
import tensorflow as tf
import math
Then, we create the training data. x_data
composed of 1000 points, and
normal noise is added to the y-coordinate of each point:
# Create noisy data
x_data = np.linspace(-10, 10, num=1000)
y_data = 0.1*x_data*np.cos(x_data) + 0.1*np.random.normal(size=1000)
print('Data created successfully')
Here is the training set:
Once our training dataset is built, we can create our network:
RELU is probably not the best choice for this application, but it works fine. ELU should provide smotther results.
# Create the model
model = keras.Sequential()
model.add(keras.layers.Dense(units = 1, activation = 'linear', input_shape=[1]))
model.add(keras.layers.Dense(units = 64, activation = 'relu'))
model.add(keras.layers.Dense(units = 64, activation = 'relu'))
model.add(keras.layers.Dense(units = 1, activation = 'linear'))
model.compile(loss='mse', optimizer="adam")
# Display the model
model.summary()
The model is compiled with the following optimization parameters:
optimizer="adam"
), more info here.loss='mse'
). More information about metrics on this page.Once the model is defined, let's train our network:
x_data
is the inputy_data
is the expected outputepochs=100
means our network will be trained 100 times with our datasetverbose=1
display progression and loss in the console. # Training
model.fit( x_data, y_data, epochs=100, verbose=1)
It should display something like (loss should decrease):
Train on 1000 samples
Epoch 1/100
1000/1000 [==============================] - 0s 321us/sample - loss: 0.2125
Epoch 2/100
1000/1000 [==============================] - 0s 49us/sample - loss: 0.1914
Epoch 3/100
1000/1000 [==============================] - 0s 50us/sample - loss: 0.1932
Epoch 4/100
1000/1000 [==============================] - 0s 60us/sample - loss: 0.1922
...
Epoch 97/100
1000/1000 [==============================] - 0s 59us/sample - loss: 0.0180
Epoch 98/100
1000/1000 [==============================] - 0s 53us/sample - loss: 0.0188
Epoch 99/100
1000/1000 [==============================] - 0s 54us/sample - loss: 0.0161
Epoch 100/100
1000/1000 [==============================] - 0s 55us/sample - loss: 0.0147
Once trainning is over, we can predict and display the output for each input:
# Compute the output
y_predicted = model.predict(x_data)
# Display the result
plt.scatter(x_data[::1], y_data[::1])
plt.plot(x_data, y_predicted, 'r', linewidth=4)
plt.grid()
plt.show()
Here is the result:
You can try this example online on Google Colaboratory