Do not learn with machine learning library TensorFlow ~ Fibonacci sequence

For those who are not familiar with machine learning but want to calculate within the range that they can understand, let's try Fibonacci sequence with TensorFlow.

Basic

Fibonacci sequence $ a_{n+2} = a_{n+1} + a_n, \\ a_1 = 1, \\ a_0 = 0 $

For example, to calculate

u = 1
v = 0
for i in range(100):
	print i, v 
	u = u + v
	v = u - v

You can do it like this.

Preparation

TensorFlow

import tensorflow as tf

Suppose you want to read with.

Variable First, let's change the variables u and v to TensorFlow variables. To do this, use Variable as follows:

u = tf.Variable(1, "int64")
v = tf.Variable(0, "int64")

I chose int64 as an overflow countermeasure, but unfortunately, an overflow occurs in Section 93. If this is omitted, it will be treated as int32.

display

It sounds like you can print it with print u, but unfortunately you won't get what you expect. TensorFlow is divided into a phase of defining variables, functions and procedures, and a phase of actually processing data.

Use tf.Session to actually process the data. Variables need to be initialized at the beginning of the session. Use tf.initialize_all_variables for this. And you can get the values of u and v by doing the following.

init = tf.initialize_all_variables()
with tf.Session() as sess:
	sess.run(init)
	print sess.run(v)

Take values and execute procedures with sess.run.

update

The rest is the value update part. Use tf.assign for assignment. Use tf.add and tf.sub for addition and subtraction. And let's define a procedure called substitution.

update_u = tf.assign(u, tf.add(u, v)) # u = u + v
update_v = tf.assign(v, tf.sub(u, v)) # v = u - v

Summary

To summarize the above story,

import tensorflow as tf

u = tf.Variable(tf.cast(1,"int64"))
v = tf.Variable(tf.cast(0,"int64"))

update_u = tf.assign(u, tf.add(u,v))
update_v = tf.assign(v, tf.sub(u,v))

init = tf.initialize_all_variables()

with tf.Session() as sess:
	sess.run(init)
	for i in range(100):
		print i, sess.run(v)
		sess.run(update_u)
		sess.run(update_v)

When I run it, somehow the CPU somehow displays a log that starts with I, but since it is Info, I don't care.

If you haven't touched TensorFlow yet, let's do something.