Windows

1. Anaconda3 Setup

Download and run the Anaconda3 installer (e.g., Anaconda3-2019.03 for Python 3.7). During installation, you can allow it to register Python in the system PATH or rely on the Anaconda Prompt later.

2. Verifying the Installation

Open a terminal (cmd or Anaconda Prompt) and check the conda version:

conda --version

List existing environments:

conda info --envs

3. Creating a TensorFlow Environment

Create a dedicated conda environment with Python 3.7:

conda create --name tf_env python=3.7

Activtae it:

conda activate tf_env

To speed up pip downloads, you can configure a mirror in C:\Users\<username>\pip\pip.ini:

[global]
index-url = https://pypi.tuna.tsinghua.edu.cn/simple
[install]
trusted-host = pypi.tuna.tsinghua.edu.cn

Install TensorFlow using pip:

pip install --upgrade tensorflow

4. Smoke Test

While the environment is active, launch Python and run:

import tensorflow as tf
msg = tf.constant('Hello, TensorFlow!')
with tf.compat.v1.Session() as sess:
    print(sess.run(msg))

If you see b'Hello, TensorFlow!', the installation is correct.

Linux

1. Choosing the Right Installer

Check the system architecture and kernel version:

uname -a

Download the matching Anaconda3 Linux installer (e.g., Anaconda3-2019.03-Linux-x86_64.sh).

2. Installing Anaconda3

Run the installer script:

bash Anaconda3-2019.03-Linux-x86_64.sh

Accept the license and choose a custom installation path, for instance /home/user/anaconda3. When asked whether to initialize conda, answer yes.

To keep the system Python (e.g., Python 2.7 used by Hadoop) unchanged, edit ~/.bashrc and comment out the entire conda initialize block. Add the following line instead:

export PATH="/usr/bin:$PATH:/home/user/anaconda3/bin:/home/user/anaconda3/condabin"

Reload the configuration with source ~/.bashrc. Confirm the system Python still points to version 2:

python --version

3. Setting Up TensorFlow

Create a new environment:

conda create -n tf_env python=3.7

Activate it:

source activate tf_env

Install TensorFlow via conda (this resolves dependencies more reliably on some Linux distributions):

conda install tensorflow -n tf_env

4. Validation

Start Python in the activated environment and test:

import tensorflow as tf
msg = tf.constant('Hello, TensorFlow!')
with tf.compat.v1.Session() as sess:
    print(sess.run(msg))

Expected output: b'Hello, TensorFlow!'.