[PYTHON] Try running tensorflow on Docker + anaconda

Introduction

Since I bought a new PC, I used Docker to build a python environment and run CNN. Since Docker is almost a beginner, I will take this opportunity to study.
As the flow this time

1. 1. Get the official anaconda3 image on Docker
2. Create a container based on the anaconda3 image
3. 3. Introduce tensorflow to anaconda3 Four. Open Jupyter Noteook from your browser Five. Increase Docker memory
4. Build a CNN using tensorflow It has become.

environment

• MacOS Catalina version 10.15.5

• Docker version 19.03.8

1. 1. Get the official anaconda3 image on Docker

anaconda is an open source platform that has all the essential packages for data science practitioners. I am grateful to use this because there is an image of anaconda3 from the official.

https://hub.docker.com/r/continuumio/anaconda3/

Now, start Docker and get the image of Anaconda3.

％ docker pull continuumio/anaconda3

Using default tag: latest
latest: Pulling from continuumio/anaconda3
68ced04f60ab: Pull complete
57047f2400d7: Pull complete
8b26dd278326: Pull complete
Digest: sha256:6502693fd278ba962af34c756ed9a9f0c3b6236a62f1e1fecb41f60c3f536d3c
Status: Downloaded newer image for continuumio/anaconda3:latest
docker.io/continuumio/anaconda3:latest

pull is a command to get a Docker image.

2. Create a container based on the anaconda3 image

Next, we will create a container.

% docker run --name anaconda -it -p 8888:8888 -v /Users/xxxx/docker/anaconda:/home continuumio/anaconda3 /bin/bash
(base) root@xxxx:/# conda list

The meaning of each command and option is as follows.

• run: Create a container.

• --name [name]: Set the name of the container.

• -it: You can operate inside the container.

• -p [Host side port number: Container port number]: Bind the container port to the host side.

• -v [Host-side directory: Container-side directory]: Host-side directory in the container Mount
  

3. 3. Introduce tensorflow to anaconda3

Let's check the packages that are pre-installed on anaconda3.

base) root@xxxx:/# conda list
# packages in environment at /opt/conda:
#
# Name                    Version                   Build  Channel
_ipyw_jlab_nb_ext_conf    0.1.0                    py37_0  
_libgcc_mutex             0.1                        main  
alabaster                 0.7.12                   py37_0  
anaconda                  2020.02                  py37_0  
anaconda-client           1.7.2                    py37_0  
anaconda-navigator        1.9.12                   py37_0  
anaconda-project          0.8.4                      py_0  
argh                      0.26.2                   py37_0  
asn1crypto                1.3.0                    py37_0  
astroid                   2.3.3                    py37_0  
astropy                   4.0              py37h7b6447c_0  
atomicwrites              1.3.0                    py37_1  
attrs                     19.3.0                     py_0  
.
.
.
yaml                      0.1.7                had09818_2  
yapf                      0.28.0                     py_0  
zeromq                    4.3.1                he6710b0_3  
zict                      1.0.0                      py_0  
zipp                      2.2.0                      py_0  
zlib                      1.2.11               h7b6447c_3  
zstd                      1.3.7                h0b5b093_0  

The docker command cannot be used because it is in the anaconda3 container. Look at the package in the conda list available on the anaconda terminal.
Since anaconda3 does not have tensorflow installed from the beginning, use conda install Install tensorflow and its peripheral packages.

base) root@xxxx:/# conda install tensorflow

Four. Open Jupyter Notebook from your browser

You can open the Jupyter Notebook, which you used to love when you were in college, from your browser.

base) root@xxxx:/# jupyter notebook --port 8888 --ip=0.0.0.0 --allow-root
.
.
.
To access the notebook, open this file in a browser:
        file:///root/.local/share/jupyter/runtime/nbserver-12-open.html
    Or copy and paste one of these URLs:
        http://81127b992594:8888/?token=22ab4d7b42e6629eb76dad08af9c8e6b1d5b59e0f0050f73
     or http://127.0.0.1:8888/?token=22ab4d7b42e6629eb76dad08af9c8e6b1d5b59e0f0050f73

Five. Increase Docker memory

Before building CNN, Docker can only use up to 2G of memory by default, so if you try to train with CNN, it will overflow. So I will change the setting so that it can be used up to 7G.

1. Click the Docker icon at the top of your desktop
2. Click Preference
3. Click the Resources tab
4. Set Memory to 7G

6. Build a CNN using tensorflow

Finally, we will build CNN. For the dataset, download the standard CIFAR10, and use the three convolution layers, the adam method for optimization, and the cross entropy error for the loss function.
I was able to import tensorflow without any problems, and the memory did not overflow, so this is the end of environment construction!

At the end

I tried touching Docker with a light feeling, but it was too deep and I was addicted to the swamp ... Introducing tensorflow to anaconda3 and opening Jupyter Notebook, which was probably quite difficult for experienced Docker users. There is no loss in doing Docker, so I would like to actively use it. For the time being, I was able to build an environment where tensorflow can be operated, so I will try various deep learning in this environment from now on.