Teaching ML

TF-Introduction

TF Playground

This is a tutorial for better understanding the concept and syntax of Tensorflow. In this tutorial, we assume you already have the basic knowledge of Python and Machine Learning. You can fill the blank lines, run the script and see what would happen according to the code. This tutorial will begin with some style advice and some basic concepts, following by some exercises of different Machine Learnig models.

Please feel free to add some chapters if you think it might be good for our group.

Agenda

Sample Code

Here is a sample Tensorflow code, play around and grab as much information as you can here. We’ll explain it in detail in later chapters. When you think the code is ready, toggle the “run cell” button in the menu bar and see what will happen.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
from __future__ import absolute_import
from __future__ import print_function
from __future__ import division
import tensorflow as tf
import numpy as np
## input
# randomly generate 100 instances with float type
x_data = np.random.rand(100).astype(np.float32)
# create a linear model: y = x * 0.1 + 0.3
y_data = x_data * 0.1 + 0.3
## build the graph(we already know w = 0.3 and b = 0.1 though, tensorflow will figure it out by its own)
# initialize uniform distributed weight vector
W = tf.Variable(tf.random_uniform([1], -1.0, 1.0))
# initialize bias to 0
b = tf.Variable(tf.zeros([1]))
# define the graph
y = W * x_data + b
# define the loss function
loss = tf.reduce_mean(tf.square(y - y_data))
# using gradient descent minimize the loss 
optimizer = tf.train.GradientDescentOptimizer(0.5)
train = optimizer.minimize(loss)
## initialization and launch the graph
# define the initialization
init = tf.initialize_all_variables()
# get a session and initialize the variables
sess = tf.Session()
sess.run(init)
## training
for step in range(201):
    sess.run(train)
    if step % 20 == 0:
        print("steps: %i, weight: %f, bias: %f" % (step, sess.run(W), sess.run(b)))
steps: 0, weight: 0.098388, bias: 0.440273
steps: 20, weight: 0.081874, bias: 0.310411
steps: 40, weight: 0.094683, bias: 0.303054
steps: 60, weight: 0.098440, bias: 0.300896
steps: 80, weight: 0.099542, bias: 0.300263
steps: 100, weight: 0.099866, bias: 0.300077
steps: 120, weight: 0.099961, bias: 0.300023
steps: 140, weight: 0.099988, bias: 0.300007
steps: 160, weight: 0.099997, bias: 0.300002
steps: 180, weight: 0.099999, bias: 0.300001
steps: 200, weight: 0.100000, bias: 0.300000

In the above sample, we defined a set of random inputs with y = 0.1 * x + 0.3 and let our linear model y = W * x + b to learn the inputs in 200 training epochs. Results are printed per 20 time steps. W and b should be quite close to 0.1 and 0.3 respectively.

Next Chapter: Style Guide