3 of 20

About myself

Felipe Caballero
Full Stack Developer
Worked for a while doing websites
Interested in Data Science
Recommended course: deeplearning.ai (uses Python)

https://github.com/caballerofelipe

https://medium.com/@felipecaballero

https://www.linkedin.com/in/felipe-caballero-74578412/

A TensorFlow very simple Linear Regression model

4 of 20

Objectives

Train a Linear Regression model
Use TensorFlow in a very simple way
Show how TensorFlow’s low level API works

A TensorFlow very simple Linear Regression model

5 of 20

WTF is Linear Regression?

We have points in a space
We need to have a good estimator for new points
We create a function to help us estimate

A TensorFlow very simple Linear Regression model

6 of 20

How to estimate a function using Linear Regression?

Define an equation form to estimate all points
The idea is to have an equation as close as possible to all existing points
Define a cost function which will tell how close the estimator is to all the points
Minimize the cost
The estimator is dependent on a and b
The cost function also is dependent on a and b

The idea of cost minimization is used a lot in Machine Learning.

A TensorFlow very simple Linear Regression model

7 of 20

Minimize cost using Gradient Descent

Gradient descent uses the derivative of a function at a certain point to travel to a lower point
It does this several times until the minimization reaches convergence
The learning rate is important as this can make Gradient Descent work in unexpected ways

Image from https://flylib.com/books/2/71/1/html/2/files/17fig06.gif

A TensorFlow very simple Linear Regression model

8 of 20

Ok, so… how do we do it?

Prepare the environment (conda)
Prepare Python to handle the model (imports)
Prepare the data (and if want also check it)
Create the TensorFlow Graph
Run the model
Check the results

A TensorFlow very simple Linear Regression model

9 of 20

Install stuff

Activate

Test

Create dir

Create file

Prepare the environment

$ conda create -n envName python=3.6 tensorflow matplotlib jupyter

$ source activate envName

(envName) $ python --version

(envName) $ mkdir linreg

(envName) $ cd linreg

(envName) $ touch main.py

A TensorFlow very simple Linear Regression model

10 of 20

Importing Libraries

# Prepare stuff

import numpy as np

import matplotlib.pyplot as plt

import tensorflow as tf

# To be used in Jupyter Notebooks

from tensorflow.python.framework import ops

A TensorFlow very simple Linear Regression model

11 of 20

Disable warning

import os

# See https://stackoverflow.com/a/47227886/1071459

# Just disables the warning, doesn't enable AVX/FMA

# To avoid tensorflow warnings, uncomment the following line

os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'

A TensorFlow very simple Linear Regression model

12 of 20

Prepare the data

# Prepare the data

x = np.array([

-7, 5, 7, 12, 2,

4, 5, 6, 19, 14,

16, 15, 11, 8, 18,

-2, 14, 17, 7, 17

])

y = np.array([

-7.56, 21.01, 29.11, 47.89, 14.11,

16.69, 29.81, 28.67, 63.94, 54.8,

49.65, 52.26, 44.86, 40.45, 70.92,

-0.37, 44.34, 64.27, 32.85, 50.14

])

A TensorFlow very simple Linear Regression model

13 of 20

Check the data

# Check the data

print('x.shape():')

print(x.shape)

print('y.shape():')

print(y.shape)

input("Press enter to continue...")

plt.scatter(x, y)

plt.show()

A TensorFlow very simple Linear Regression model

14 of 20

The TensorFlow part — Config

# TensorFlow Model

# Config

num_epochs = 1000

learning_rate = 0.001

# /Config

A TensorFlow very simple Linear Regression model

15 of 20

The TensorFlow part — If using Jupyter

# This is necessary if the code is run inside Jupyter

ops.reset_default_graph()

A TensorFlow very simple Linear Regression model

16 of 20

The TensorFlow part — The graph

# Creating the graph

X = tf.placeholder(tf.float32, name='X')

Y = tf.placeholder(tf.float32, name='Y')

a = tf.get_variable('a', initializer=0.)

b = tf.get_variable('b', initializer=0.)

h = a * X + b

cost = tf.reduce_mean( (h - Y)**2 )

optimizer = tf.train.GradientDescentOptimizer(

learning_rate=learning_rate

).minimize(cost)

init = tf.global_variables_initializer()

A TensorFlow very simple Linear Regression model

17 of 20

Run the model — Store result in vars

# Running the Model

found_a = 0

found_b = 0

A TensorFlow very simple Linear Regression model

18 of 20

Run the model — The actual run

with tf.Session() as sess:

sess.run(init)

for epoch in range(num_epochs):

_, costValue = sess.run(

[optimizer, cost],

feed_dict={

X: x,

Y: y,

}

)

found_a = a.eval()

found_b = b.eval()

if epoch % (num_epochs/10) == 0: # Every 10 percent

print("... epoch: " + str(epoch))

print(f"cost[{str(costValue)}] / a[{str(a.eval())}] / b[{str(b.eval())}]")

A TensorFlow very simple Linear Regression model

19 of 20

Visualize the results

# Seeing the obtained values in a plot

xrange = np.linspace(-10, 30, 2)

# Plot points

plt.plot(x, y, 'ro')

# Plot resulting function

plt.plot(xrange, xrange * found_a + found_b, 'b')

plt.show()

A TensorFlow very simple Linear Regression model

20 of 20

Possible growth — …to enhance the code

We could:

Enhance the model, add a square element in the equation for instance
Save the data, save found values to disk
Save iterations, save different values in time

A TensorFlow very simple Linear Regression model