This paper introduces the detailed TensorFlow

installed on windows with a simple example, for everyone to share, as follows:



version currently can be installed in Ubuntu, Mac OS, Windows, CPU: GPU version of >

version of the
installation: pip, Anaconda


  1. Windows now supports python3.5.x
  2. gpu version cudnn5.1


2017/3/4 installation progress progress:
Anaconda 4.3 (corresponding to python3.6) is being installed, delete, all
2017/3/5 schedule:
Anaconda 4.3 get
(corresponding to python3.6) Anaconda python3.5.2get

in the train of thought look at the others to teach During the journey, there are some nouns that they have never seen, and they are daunting.
so the claim starts from the noun interpretation.
then tells the installation and simple examples of TensorFlow.
as his own note,
also wants to see this tutorial like my little white.


CUDA (Compute Unified Device Architecture) is a computing platform launched by the developer NVIDIA. It is a kind of general CUDA launched by the NVIDIA parallel computing architecture, this architecture enables GPU to solve complex computational problems. It contains the CUDA instruction set architecture (ISA) and the parallel computing engine within the GPU. Developers can now use the C language to write a program for the CUDA architecture, C is a high-level programming language most widely used. Write the program so they can be in support of the CUDA processor to ultra high performance. CUDA3.0 has started to support C++ and FORTRAN. The
computing industry is developing from "central processing" using only CPU to the "collaborative process" used by CPU and GPU. In order to build a new computing model, NVIDIA (NVIDIA TM) invented CUDA (Compute Unified Device Architecture, Compute Unified Device Architecture) this programming model is to make full use of the advantages in the application of CPU and GPU respectively. Now, the architecture has been applied to the GeForce (TM, ION (sperm) Yiyang, Quadro and Tesla TM) GPU (graphics processing unit).
comes from Baidu encyclopedia.

(so my A card is useless)


Anaconda is a leading open data science platform supported by Python. The open source version of Anaconda is a high-performance distribution version of Python and R, including more than 100 popular Python, R and Scala packages for data science.
comes from Anaconda official download page

, see the specific use of the "", "ah", "the right way", the official tutorial is simple and easy to understand.

Anaconda preliminary study

0. download Anaconda package: Anaconda

official download address is Anaconda4.3.0For Windows 64bit I download (built-in python3.6)


well installed, has been the next step.

1. check Anaconda: conda --version

(hee hee, the first step in successful, happy)

2. detection of the current installation environment: conda info --envs


" > (only one! Not afraid, go on!)

3. what are the current version of the python conda search --full-name python can be installed:

(many, to which? Of course is python3.5

) and

4. create --name to install different versions of the python:conda tensorflow python=3.5

(system will automatically select a 3.5.x version of the conjecture into python=3.5 version,

(hee hee! OK! A step closer to success!)

activate tensorflow

6. named tensorflow environment has been successfully added: conda info --envs

7. inspection Python version: python --version

" left "> (happy ^ ^ ~)

8. deactivate

to exit the current environment:

(small cap fell)

9. activate tensorflow

" left "> want to switch to activate ~

which environment which

this article since the installation of tensorflow, of course, avtivate tensorflow!

little goblin! I'm coming.

PS: want to know more about Anaconda official tutorial , easy to understand good! Do not search online tutorials, no official tutorials look refreshing!

is the

TensorFlow installation to install tensorflow in a native windows system, using


installation, the installation of the CPU Version (well, as the AMD

Anaconda graphics card, break out crying) (with python3.6)

in Anaconda with the following python3.5.2

is the protagonist of today!

1. (crackling) according to the official website of the
pip install --ignore-installed --upgrade


(well, the first is wrong, then this! I don't know what to do (at o on

2.) /~~) another try: pip install tensorflow

< p> (highlight, be like this! I'm a AMD card. It's not the same!)

3. tensorflow:

confirmed the successful installation of trial and error: type directly in the CMD Python tensorflow as tf

import, and then type "text-align: center"

the correct attempt: into the Anaconda Prompt-python, into the installation of the environment called tensorflow (we do python3.5.2 remember? ~), type python, and then type import tensorflow as tf

Anaconda Prompt-python:

"text-align: center" >

Anaconda Navigator (open the start menu -> Anaconda; 3-> Anaconda Navigator), a Spyder play, click on the Spyder below the "Install", "Launch" becomes installed, click to get in.

in Spyder of tensorflow


output: </p>

(hey hey hey ha ha ha I feel as if I succeed!!! What about you?


installation routines to put up a new thing, we put it together!

concept what to run through the first small program to see again!

to find a sense of achievement in order to continue!

example source: MINIST For ML Beginners


  1. 55000 data sets: the training set, 10000 test sets, 5000 sets of
  2. verification of each image was 28pixels*28pixels
 # Code: from tensorflow.examples.tutorials.mnist import input_data data set MNIST = input_data.read_data_sets ("MNIST_data/" one_hot=True import tensorflow as TF) # input image data placeholder x = tf.placeholder (tf.float32, None 784]) # weights and deviation W = tf.Variable (tf.zeros ([784, 10]) = tf.Variable (tf.zeros) B ([10])) # using softmax model y = tf.nn.softmax (tf.matmul (x, W) + B the cost function #) = tf.placeholder (tf.float32, y_ for [None, 10]) # cross entropy evaluation cost cros S_entropy = tf.reduce_mean (-tf.reduce_sum (y_ * tf.log (y), reduction_indices=[1]) # use)

This concludes the body part